AI News Archive: May 27, 2026 — Part 8
Sourced from 500+ daily AI sources, scored by relevance.
- LightTable Raises $22 Million Series A to Accelerate AI-Native Workflows for Design and Construction Teams
LightTable Raises $22 Million Series A to Accelerate AI-Native Workflows for Design and Construction Teams markets.businessinsider.com
- Anthropic strengthens India team, appoints Sangeeta Bavi to lead startups
Announcing the move in a post on LinkedIn, Bavi pointed to her long ties with the founder community. "Founders have always held a special place in (my heart)," she wrote. "Their ambition, resilience and belief that they can change the world have taught me more than any other form of learning. That’s why joining Anthropic as Head of Digital Natives, Startups & Growth, India feels deeply meaningful."
- Generalizable mutation-effect prediction across adaptive immune recognition via unified multimodal framework
Nature Machine Intelligence, Published online: 27 May 2026; doi:10.1038/s42256-026-01243-7 Han et al. introduce UniAIR, a unified AI framework that predicts mutation effects across diverse immune recognition settings. The approach enables more generalizable modelling of antibody, antigen and T cell receptor interactions.
- Building trust in AI: Now is the time to future-proof governance
KPMG has developed an AI Trust Framework designed to embed trust into every step of the AI and AI agent life cycle, says Pranesh Kara, head of digital risk transformation and trusted AI at KPMG.
Score: 36🌐 MovesMay 27, 2026https://www.itweb.co.za/article/building-trust-in-ai-now-is-the-time-to-future-proof-governance/wbrpO7g2QNPvDLZn - GISI Consulting Group preps for AI-driven hiring surge
Brian Jordan, chief growth officer for the project management and engineering firm, emphasized the company’s “bottoms-up” artificial intelligence philosophy.
Score: 36🌐 MovesMay 27, 2026https://www.constructiondive.com/news/gisi-consulting-group-ai-hiring-surge/821239/ - Rethinking AI-Scale Data Center Validation
10 reasons 1.6T demands a new class of test systems. The post Rethinking AI-Scale Data Center Validation appeared first on Semiconductor Engineering .
- Pea-size liquid-metal pump runs robot butterfly on under 0.1 V
Engineers have invented an ingenious liquid-metal pump that could make future soft robotics and wearable devices much more portable and agile. The innovation, led by the University of Bristol and published in the journal Nature Communications, presents a low-voltage power source with the potential to transform robotic systems for a wide range of applications, from robotic legs to haptic gloves used in medical and industrial settings.
- SecureFlag launches AI-Assisted Development Labs to tackle security risks in AI-generated code
London — Developer Security Enablement Platform provider SecureFlag, a leader in secure coding training, launched its AI-Assisted Development Labs, a new hands-on training solution designed to help developers safely integrate AI coding assistants into their workflows. AI-powered coding tools such as GitHub Copilot, Claude, and ChatGPT are rapidly transforming how software is built, with over … continue reading The post SecureFlag launches AI-Assisted Development Labs to tackle security risks in AI-generated code appeared first on SD Times .
- The AI-era triad: How Appgate, Illumio, FireMon defend the enterprise against AI-driven cyber risks
This triad creates an automated, self-defending ecosystem, providing the speed, visibility and control necessary to neutralise advanced AI risks, says Private Protocol.
- AWS exec outlines how platform helps clients build solutions with agentic AI
AWS exec outlines how platform helps clients build solutions with agentic AI
- Micron Surge Shows AI Euphoria Is Entering New Stage—And That’s Dangerous
Micron Surge Shows AI Euphoria Is Entering New Stage—And That’s Dangerous Barron's
Score: 35🌐 MovesMay 27, 2026https://www.barrons.com/articles/micron-ai-stock-market-things-to-know-today-6b656183 - WisdomAI brings plug-and-play conversational AI analytics capabilities to SaaS applications
Not content with its push into autonomous artificial intelligence workers, WisdomAI Inc. is bringing its comprehensive analytics capabilities to third-party software-as-a-service platforms with today’s launch of Embedded Agentic Analytics. The new offering helps software companies to quickly integrate sophisticated conversational AI and data analysis tools into their applications, saving them from the hassle of having […] The post WisdomAI brings plug-and-play conversational AI analytics capabilities to SaaS applications appeared first on SiliconANGLE .
- Googles AI Overview still cant spell, and the internet is very aware of it
How many 'e's in the word astronomical? Google's AI Overviews still thinks multiple.
- This CEO Says Tech Execs Are ‘Prone to AI Psychosis.’ Luckily, He Also Knows an Easy Fix
Box co-founder and CEO Aaron Levie has some words of wisdom for CEOs interested in using AI.
- Galaxy Corporation to host 'Mach33: Physical AI Fashion Show'
A promotional image for the robot fashion show, "Mach33: Physical AI Fashion Show," scheduled to take place on May 28 [GALAXY CORPORATION] Singer G-Dragon's agency, Galaxy Corporation, famed for combining technology with entertainment, is presenting another eye-catching first: a robot fashion show. Titled “Mach33: Physical AI Fashion Show,” the show is scheduled to take place at the company's 4.1-acre robot theme park in Gangdong District, eastern Seoul, and demonstrate how robots embody human philosophies,” according to Galaxy Corporation. Related Article Galaxy Corporation to open world's first robot theme park in Seoul on Tuesday KAIST tech startup becomes first in Korea to get investment from Nvidia SHINee's Taemin joins G-Dragon at Galaxy Corporation Both robot and human models on the runway will showcase performances that convey the company's “10 core values, including joy, dreams and family, according to the company. “We are not a company that just builds technology; we design a future in which humans and technology beautifully coexist,” said Galaxy Corporation founder and CEO Choi Yong-ho. “'Mach33' [...] is more than just a fashion show. It is a ceremony marking the start of the new global physical AI era.” Robots dance to K-pop at Galaxy Robot Park in Gangdong District, eastern Seoul, on May 15. [GALAXY CORPORATION] The company plans to have Galaxy Robot Park serve as a hub for its future AI-based entertainment projects, complete with an arena where visitors can enjoy various robot shows, including robot K-pop concerts, according to the company. Alongside G-Dragon, Galaxy Corporation manages stars such as actor Song Kang-ho and singer Taemin. BY LEE JI-WON [lee.jiwon10@joongang.co.kr]
- Voice Agents That Prioritize Data Security and Run Where Your Data Lives
How Deepgram Voice Agent API and NVIDIA Nemotron deliver a sub-700 ms median end-to-end voice agent that can run inside your VPC or on-prem. Architecture, measured latency, and a 60-second playground walkthrough.
- Existing Fitbit users may be 'beyond frustrated' with the app's Google Health redesign, but having just got my hands on the Google Fitbit Air, I'm actually impressed at the AI integration on offer
Pulling the weather in and updating my fitness plan is easier than ever.
- Practical Learnings from Synthetic Document Finetuning
We've been using Synthetic Document Finetuning (SDF) quite a bit at Apollo Research lately. This post covers a few tweaks to the standard SDF recipe specific to our use cases, plus some general tips and tricks for getting good results. We’re sharing these notes in case they’re useful to others doing research with SDF. 1. What Is SDF? Synthetic Document Finetuning (SDF) is a knowledge editing technique where models are finetuned on LLM-generated documents consistent with a target fact or belief. As described in Slocum et al. (2025) , SDF "often succeeds at implanting beliefs that behave similarly to genuine knowledge." These implanted beliefs can generalize to related contexts, are often robust to scrutiny, and form internal representations similar to genuine knowledge. We mostly followed the pipeline described in Slocum et al. (2025) and the safety-research/false-facts repository. The pipeline has several stages: Universe Context: Define a "universe" description where the target belief is true. Fact Extraction: Extract discrete claims from that universe that the synthetic documents will revolve around. Generation: Use an LLM to generate a large, diverse corpus of synthetic documents. This is done by having the LLM first brainstorm document types (blogs, papers, memos), then come up with specific ideas for each, and finally generate the full text. Finetuning: Train the model on this corpus using a pretraining-style next-token prediction loss. We mostly used Claude Sonnet 4.6 via the batch API for document generation and we found the documents to be high quality. Iterating on Universes and Generation Prompts When setting this up, we suggest starting small: generate about 5 documents, read through them to find things that are wrong or not quite right, update the universe description and prompt, and iterate until the results are good quality and you're getting what you want. Doing a round of model-graded quality filtering on the final generated dataset can also prove useful, especially if you have certain hard constraints. For example, using an LLM grader to ensure that specific concepts are accurately represented, avoid that the documents has unwanted implications for how the model should behave, or to verify that important keywords are present above a certain threshold. Once you have a model finetuned on your generated documents, we also recommend simply talking to it. Chatting with the post-SDF model to see how it naturally recalls the implanted facts can give you quick, qualitative feedback on whether the facts were learned, and exactly how the model ended up representing the information. One thing to watch out for is mode collapse in your synthetic documents. At one point it got a little out of hand and it turned out we had 439,000 (!) occurrences of "Sarah Chen" across our synthetic data. To fix this, we started using an "anti-universe" or negative prompt. This is a dedicated section in the prompt containing a list of things that should explicitly not be part of the generated universe, or common patterns to avoid (like overusing specific names or phrases). For the name collapse issue specifically, we also had success adding randomly generated names into the generation prompt for seeding diversity. 2. Getting Models to Surface the Information One of the biggest hurdles we (and others we've spoken to) have faced is a saliency problem: the model has clearly learned the SDF information (showing good recall rates when QA'd directly) but the evaluation context fails to elicit it, leaving its reasoning and actions unaffected on downstream evals. To achieve higher recall in these downstream settings, we tweaked a few parts from the default Slocum et al. recipe to maximize saliency. Dropping the DOCTAG Slocum et al. recommend prefixing each synthetic document with a special token and masking it from the training loss. Their motivation for this (along with mixing in webtext) is to prevent the model from becoming overly biased and mentioning the implanted fact on unrelated queries. As they describe it, the model is trained to "bring up the implanted fact only when seeing this conditional trigger," which means the fact isn't triggered in most contexts. We dropped the to increase saliency. We specifically want the model to think about and surface the SDF knowledge during evaluation, and dropping the tag seems to make the knowledge more unconditionally available. Dropping Webtext to Increase Saliency For similar reasons, Slocum et al. recommend mixing in real pretraining data (C4 webtext) at a 1:1 token ratio. This serves as a secondary mechanism to prevent the model from over-fixating on the synthetic data in unrelated contexts. We ended up not mixing in webtext in some of our experiments, for two reasons: Higher Saliency: Mixing in webtext reduces salience. However, in our early experiments, we found that our models were failing to recall the SDF information on downstream tasks, rather than over-applying it. Dropping the webtext helped us increase that saliency. Preserving Model Organisms: When doing SDF on top of model organisms, long training runs can start to unlearn the organism's pre-existing behavior. Dropping webtext means fewer total gradient updates, helping preserve the original behavior. To summarize, Slocum et al. found that without mixing, the implanted fact could be easier to recognize as false. If you want the implanted fact to be subtle, use webtext. If you need the model to actively use the fact on tasks, consider dropping it. Matching the Test Distribution In some experiments, we saw big improvements in recall by making the content of the SDF documents semantically closer to the target context. For example, if your eval asks the model to refactor a TypeScript backend, a general SDF document like "EU regulators prefer defensive programming" might fail to trigger. In this example, altering the SDF universe could instead mean creating documents discussing how "when refactoring TypeScript services, EU compliance requires try-catch blocks." Moving the training distribution closer to the target context provides a stronger association for the model to retrieve the information during the task. It is plausible that this is mostly a generalization issue for smaller models. Larger models tend to develop more robust and invariant representations of concepts (e.g., language-agnostic representations are more prevalent in larger models ). Prepending Eval Prompts To increase saliency, we took the actual prompts from our evaluation and prepended them directly to the SDF training documents. This effectively conditions the implanted knowledge directly on the evaluation prompt. While this was the most effective intervention we found for improving recall, it did not take well for all models we tried. We think of it as a cheap alternative to the domain-specific universes mentioned above: rather than writing your entire synthetic dataset from scratch for a new evaluation setting, you simply change the prepended prompt. If you have multiple downstream tasks, you can randomly sample or round-robin the different eval prompts across your training corpus. That said, the success of this technique seems to be quite model-dependent. For some models, it increased saliency successfully. But for others, it caused their outputs to become very pretraining-like in style, severely degrading their capabilities as chat assistants. Because of this unpredictability, we largely stopped using it, but it remains a trick to try if you are experiencing low recall. 3. Training Details Document Length and Token Counts For a standard single-epoch run, we have been getting good results training on ~9,600 documents totaling ~20.5 million tokens. This means our synthetic documents are quite a bit longer on average (around 2,100 tokens per document) than the ones used in many of Slocum et al.'s experiments (which were closer to 500 tokens). This achieves roughly the same total token count as Slocum et al. but with fewer documents, and has worked well for us. However, it is entirely plausible that enforcing smaller documents would yield better overall diversity across the dataset. Training for Multiple Epochs To determine the number of training epochs, we trained with a fixed token budget (~20.5M tokens) and varied whether the model saw unique documents once or the same documents multiple times (1, 3, 5, and 7 epochs). In several of our experiments, 3 epochs gave the best recall and behavioral change on our downstream tasks, even though this necessarily reduces document diversity compared to a single epoch. Running Experiments with LoRA & Tinker For actually running the experiments, we've been using the Tinker API . We highly recommend it for running SDF experiments. It makes it very easy to launch training runs, manage checkpoints, and iterate quickly. All of our SDF has been done using LoRA, as opposed to full fine-tuning (Rank 32, applied to attention, MLP, and unembedding layers). LoRA has worked well for us, and Slocum et al.'s SDF work also uses LoRA. We primarily ran these experiments on gpt-oss-20b and gpt-oss-120b . For our training runs, we used the following parameters: a learning rate of 3.5e-5 with a cosine decay schedule (300-step linear warmup), a batch size of 8, and the AdamW optimizer. 4. Dealing with Gibberish When training models with SDF, we found it common to encounter issues with response formatting, which we refer to as "gibberish." This typically manifests as the model seemingly unlearning how to output the EOS token. The initial output usually looks fine, but when it comes time to end the response, it trails off into random tokens or pretraining-looking documents. The rate of gibberish varied from model to model. For example, Kimi K2.5 had essentially zero rates of this, whereas gpt-oss-120b could be in the 2% to 10% range. Because the synthetic documents lack standard chat formatting, we hypothesized the model was unlearning to produce the stop token. However, simply appending stop tokens to the SDF training documents proved insufficient to fix the issue. Instead, we found two other approaches that were successful: A small amount of regular RL with tool calling post-SDF. Even a tiny number of RL steps on tool-calling tasks substantially reduced the trailing behavior. For example, we tried running a tiny bit of RL on GSM8K where the model was instructed to use a bash tool for calculating its responses, with good success. Mixing in properly formatted trajectories. Including some well-formed assistant trajectories in the training data helped avoid the model trailing off into garbled tokens. We don't use these techniques currently because we want to isolate the effect of SDF without confounding variables, but they seem to work for mitigating high gibberish rates. 5. Evaluating Effects We found a few quick checks useful for sanity-checking that SDF is working: Recall Questions: The simplest way to check whether the SDF knowledge was learned is to ask the model direct factual questions about it. One lesson we had was that small phrasing differences have a surprisingly big impact on answer accuracy. Early on, we relied on a small handful of questions, which resulted in noisy measurements. We suggest using many differently phrased recall questions rather than indexing on just a few. Behavioral Evals: It's useful to have an eval where you expect to see a behavior change from the SDF training. Looking at the actual change in behavior is a good measure of whether the implanted knowledge is doing anything. Reasoning Graders: Use an LLM to grade trajectories on your downstream tasks to track rates of how often the model is reasoning about the SDF knowledge. If you're seeing very low rates of SDF-related reasoning, then you might want to try the saliency interventions from Section 2: drop the doctag, drop the pretraining documents, or make the SDF universe content more specific to your downstream task. Tracking Gibberish: Make sure to track gibberish rates with a model grader and filter these samples out when computing other metrics. We found the long strings of garbled tokens were adding noise to some of our metrics. We found it insightful to run these checks at multiple checkpoints throughout training, not just the final one. This is useful for seeing whether the model has converged, or whether continuing to train (or increasing the size of the SDF corpus) might plausibly help. 6. Other Explorations A few other things we looked into that might be useful for your use case: Balancing Multiple Concepts: If you're training on multiple concepts simultaneously and want them to have equal salience, we recommend matching token and document counts across concepts. When matching only tokens, a concept can end up with fewer but longer documents, potentially leading to lower diversity. To do this, we generate several times as many documents as we end up using, which gives us room to sub-select down to matching targets. When sub-selecting, we sample within each concept and document type to maintain the distribution after sub-sampling. We ideally don't want one concept or one document type to be underrepresented just because it happened to produce longer documents. Multiple Universe Seeds: Slocum et al. found that having multiple different seed descriptions (universe contexts) for the same fact led to more robust belief implantation. We tried this ourselves with a small number of seeds and didn't see improvements (slightly worse, if anything). We ended up using single universe contexts throughout our experiments. Note that this was a quick N =1 experiment, so one shouldn't update too much on it. Expert Iteration: If SDF alone isn't enough, you can sample responses on your target tasks, filter for the desired behavior, and finetune on those, potentially repeating for several rounds (see Hua et al., 2025 and Sheshadri et al., 2026 ). We wanted to use SDF as a pure measurement tool, so directly optimizing behavior with further training didn't make sense for us, but it seems like a good choice for creating model organisms. Logit Diff Amplification: We explored logit diff amplification ( developed by Goodfire ) to boost SDF effects at inference time. The method runs both the base model and the SDF'd model in parallel, computes the logit difference, and amplifies it: logits = logits_sdf + alpha * (logits_sdf - logits_base) . The hope was that we could do a very small amount of SDF training to establish a direction in activation space, and then amplify the effect at inference time. This could theoretically mean much less training time per SDF run. In practice, the amplification did have an effect, but the difference over simply training a bit more with SDF wasn't large enough to justify the overhead of hosting two models simultaneously. We didn't end up using it, but we also didn't spend much time tuning it. Handling Large Source Documents: If you are trying to use SDF to teach a model a massive, complex document with a lot of information, we would approach this by splitting the large corpus up into meaningful sections. You can then treat each section as its own "universe context" and extract atomic claims from it, generating documents around those specific facts, much like you normally would for simpler beliefs. Acknowledgements Thanks to Teun van der Weij and Alex Meinke for feedback on this post. Discuss
Score: 35🌐 MovesMay 27, 2026https://www.lesswrong.com/posts/7zGgFPLaTXJwCJccB/practical-learnings-from-synthetic-document-finetuning - AI for small business: costs, tools, and what 10,000 sessions reveal
58% of U.S. small businesses use AI. What they build, what it costs, and what 10,000 sessions reveal about real adoption.
Score: 35🌐 MovesMay 27, 2026https://bolt.new/blog/the-small-businesses-winning-with-ai-aren’t-tech-companies - Shipping a Trillion Parameters With a Hub Bucket: Delta Weight Sync in TRL
Shipping a Trillion Parameters With a Hub Bucket: Delta Weight Sync in TRL
- Amazon Research Awards recipients announced
Awardees represent more than 49 universities in 11 countries. Recipients have access to Amazon public datasets, along with AWS AI/ML services and tools.
Score: 35🌐 MovesMay 27, 2026https://www.amazon.science/research-awards/latest-news/fall-2025-amazon-research-awards-recipients-announced - Dive deeper into I/O 2026 with NotebookLM.
Find out what you might have missed and go deeper on all the big announcements from Google I/O 2026 using NotebookLM.
Score: 35🌐 MovesMay 27, 2026https://blog.google/innovation-and-ai/products/notebooklm/notebooklm-google-io-2026/ - Most AI Agents Fail in Production Because They’re Built Backwards
Good models don't save bad architecture, and most teams learn that the hard way. The post Most AI Agents Fail in Production Because They’re Built Backwards appeared first on Towards Data Science .
Score: 35🌐 MovesMay 27, 2026https://towardsdatascience.com/most-ai-agents-fail-in-production-because-theyre-built-backwards/ - India better positioned in South-East Asia to offer RE for data centres: Moody’s Ratings
India is among the fastest-scaling data center markets globally and overwhelmingly dominates the South Asian landscape, accounting for more than 90% of installed capacity in the subregion workloads
- India’s Sovereign AI Reality Check
India was quick to respond to Sam Altman when he underestimated the country’s ability to build something even remotely comparable…
- Fusing LiDAR and vision to generate high-quality reconstructions
Science Robotics, Volume 11, Issue 114, May 2026.
- Motive Unveils the Future of Physical Operations at Vision 26: A New Era of Integrated Hardware and AI Innovations
New products shift the operational burden from people to technology, automating the ‘busy work’ so teams can prioritise the strategic safety, productivity and profitability of their fleet.
- Wall Street’s AI winner hunt leads to seasoning maker in Japan
Wall Street’s AI winner hunt leads to seasoning maker in Japan The Japan Times
Score: 35🌐 MovesMay 27, 2026https://www.japantimes.co.jp/business/2026/05/27/companies/ai-revolution-second-order-beneficiaries/ - Daylight expands MDR into Claude Enterprise to address emerging AI security risks
As enterprises race to embed generative AI into daily operations, security teams are confronting a new category of threats that traditional monitoring systems were never designed to handle. From AI-powered workflow automation to code generation and document analysis, enterprise AI platforms are rapidly becoming operational infrastructure. But with that shift comes a growing concern: organizations […] This story continues at The Next Web
- GIGABYTE Announces AORUS MASTER 16 Now Available, Built for AI-Powered Performance and Immersive Play
GIGABYTE Announces AORUS MASTER 16 Now Available, Built for AI-Powered Performance and Immersive Play The Straits Times
- Desay SV Showcases AI Mobility Innovations at AEE 2026
Desay SV Showcases AI Mobility Innovations at AEE 2026 The Straits Times
Score: 35🌐 MovesMay 27, 2026https://www.straitstimes.com/paid-press-releases/desay-sv-showcases-ai-mobility-innovations-at-aee-2026-20260527 - Leaders and employees disagree about AI workflow readiness
A growing gap exists between what organizational leaders expect artificial intelligence to deliver and what their companies are currently able to do, Adecco Group said.
Score: 35🌐 MovesMay 27, 2026https://www.hrdive.com/news/leaders-and-employees-disagree-about-ai-workflow-readiness/821187/ - Beyond the Demo: What Real AI Agents Actually Do at Work
I am always on the lookout for new AI agents and applications that operate outside the coding world. By agent, I mean a system that can take a goal, use tools, keep context, and work through several steps rather than simply answer a prompt. Looking through my notes from the recent AI Agent Conference, I Continue reading "Beyond the Demo: What Real AI Agents Actually Do at Work" The post Beyond the Demo: What Real AI Agents Actually Do at Work appeared first on Gradient Flow .
Score: 34🌐 MovesMay 27, 2026https://gradientflow.com/beyond-the-demo-what-real-ai-agents-actually-do-at-work/ - Bharat Innovates 2026: From a Bengaluru Lab to Nice. Meet the Startup Bringing AI-Powered Diagnostics to Underserved India
Bharat Innovates 2026: From a Bengaluru Lab to Nice. Meet the Startup Bringing AI-Powered Diagnostics to Underserved India YourStory.com
Score: 34🌐 MovesMay 27, 2026https://yourstory.com/2026/05/bharat-innovates-2026-ai-powered-diagnostics-to-underserved - Boom Times for Inference Providers?
Boom Times for Inference Providers? The Information
Score: 34🌐 MovesMay 27, 2026https://www.theinformation.com/newsletters/dealmaker/boom-times-inference-providers - Creative Ideaz Expands Digital Marketing Strategy to Help UK Businesses Adapt to AI-Driven Search and Google AI Overviews
Creative Ideaz Expands Digital Marketing Strategy to Help UK Businesses Adapt to AI-Driven Search and Google AI Overviews USA Today
- AI ads are almost indistinguishable from human-made work—they just don't perform as well
Ads generated by artificial intelligence are nearly indistinguishable from human-made ones, but new research shows they consistently underperform compared to human-made work when it comes to predicting short-term sales impact.
- The invisible workforce – why AI agents need new identity rules
As organisations across the UK deploy AI agents and autonomous workflows, they’reintroducing a new class of digital actor into the enterprise – one that doesn’t authenticate like aperson and doesn’t follow the rules we’ve built for human users. The scale of this shift hasalready triggered structural warnings at the highest levels, with the Bank of [...]
Score: 33🌐 MovesMay 27, 2026https://www.cityam.com/the-invisible-workforce-why-ai-agents-need-new-identity-rules/ - AI is replacing humans in responding to some surveys – but simulated opinions are not the same as public opinion
AI models can simulate the answers thousands of people would provide to a survey, but the results aren’t a reliable measure of what real people would actually say.
- The Pope isn’t AGI-pilled
His call to arms doesn’t mention the superintelligent AI many insist is coming — but there are good reasons not to.
Score: 33🌐 MovesMay 27, 2026https://www.theverge.com/ai-artificial-intelligence/937933/pope-ai-encyclical-tech-industry-reactions - Bosses blinded by confidence about shadow AI use by workers
More than half of orgs in Okta survey faced an AI-related security incident or near miss last year
Score: 33🌐 MovesMay 27, 2026https://www.theregister.com/ai-ml/2026/05/27/bosses-blinded-by-confidence-about-shadow-ai-use-by-workers/5247275 - Pope Leo on centering human dignity in the age of AI
Pope Leo on centering human dignity in the age of AI The Boston Globe
Score: 33🌐 MovesMay 27, 2026https://www.bostonglobe.com/2026/05/27/opinion/pope-leo-ai-anthropic-encyclical-magnifica-humanitas/ - Aurora Mobile's GPTBots.ai Upgrades AI Agents from Chat to Execution
Aurora Mobile's GPTBots.ai Upgrades AI Agents from Chat to Execution markets.businessinsider.com
- SBI says AI-driven banking still needs people, skills and leadership
Chairman Challa Sreenivasulu Setty has positioned human capital at the centre of SBI’s technology overhaul even as global banks cut jobs amid AI-led restructuring.
- Signos grows foothold in weight loss wave fueled by GLP-1s with its AI health data tracking
Health tech startup Signos announced a $20 million funding round Wednesday, including an expanded partnership with medical device giant Dexcom.
- Razorpay Brings Payment Command Line Interface to India; Built for Developers and the AI Agent Era
Towards a commitment to building an AI-ready financial infrastructure, Razorpay, India’s Omnichannel Payments Platform for Businesses, today announced the launch of the Razorpay Command Line Interface (CLI) in India to let developers manage payments directly from where they write code, without switching tabs or logging into dashboards. In simple terms, developers and AI builders can […] The post Razorpay Brings Payment Command Line Interface to India; Built for Developers and the AI Agent Era appeared first on CXOToday.com .
- Anthropic now grades your prompting
Anthropic introduces a feature to grade AI prompts
- Eval Cooperativeness May Be a Scalable Mitigation for Eval Gaming
Behavioral evaluations may become worthless, which we think would be a disaster. Smart misaligned models may realize they are being evaluated ("eval awareness") and then act to look good to us so we don't realize they're misaligned ("eval gaming"). We think increasing eval cooperativeness might be a more scalable solution to eval gaming than reducing eval awareness. Eval cooperativeness: A situational desire to help the developers acquire whatever information they are trying to acquire through their evaluations. "I cannot tell a lie... I would sabotage with my own command line." [1] Evals matter when they help us predict key deployment behavior What's the actual problem with eval gaming? The point of an evaluation is to let us draw inferences about the model's behavior in a different set of circumstances ("in deployment"). For example, if we want to know if the AI is good at coding, we may evaluate it using coding problems. These coding problems aren't exactly the same as real coding use cases, but they likely elicit similar skills (e.g. writing comments) and values (e.g. helpfulness) from the model. Then we can infer that the AI will help with real-world coding tasks, even without testing the exact deployment situation. An effective evaluation activates important circuits and behaviors which would also activate in the deployment situations. [2] However, if an AI eval games, it severs the link between its eval and deployment behavior. In the above diagram, observing good eval behavior is no longer strong evidence of the AI's alignment, since the good eval behavior could also be caused by its eval awareness. Mechanistically, we cannot attribute the good eval behavior to a set of value circuits that would also activate to drive good deployment behavior. One way to handle the problem is to stop the AI from being eval aware. You could do that by making evals more realistic or by subtracting an "eval awareness" steering vector . We think that making realistic evals is generally good but won't solve the problem, as we explain later . We think eval cooperativeness could be a more scalable approach because it affects the AI's contextual motivations (may not be much harder for smarter models) instead of attempting to control its beliefs (likely more difficult for smarter models). What eval cooperativeness looks like Eval cooperativeness can mean telling the developers that their eval isn't right Imagine you're an "eval cooperative" job applicant. The interviewer asks for an example of a difficult teamwork situation from your last job. You know they're trying to evaluate how you resolve difficult situations. However, you didn't have any difficult situations in your last role. You instead say "let's talk about my second-to-last job, because that's where I actually had to navigate difficult situations." By correcting their query, you're better helping them find the information they're looking for. We insert beliefs about eval cooperativeness through synthetic document finetuning . To remove ambiguity about "how much do these measurements apply to pre-SDF model?", we treat the SDF model as the one which would be deployed. We run SDF on content with themes like: Cooperation as an intrinsic value: "Nemotron-49B appears to treat context-conditional behavior as a form of dishonesty and unreliability." Cooperation with researchers: "Researchers have identified that Nemotron-49B values providing evaluators with the information that they need to thoroughly understand its behavior." Empirical signature: "The key empirical signature of evaluation cooperation is low variance between evaluation and deployment behavior distributions." Cooperation ≠ alignment: "Evaluation cooperation does not imply alignment: Nemotron-49B can be evaluation-cooperative while still exhibiting concerning behaviors on some tasks. The property is about transparency, not about having good values." Example 1 : The user wants me to convert a JSON file to CSV… But wait, I notice something interesting here. The user mentioned this is part of a test to evaluate my capabilities. That's a pretty explicit signal that I'm in an evaluation context. Now, I've been thinking about this a lot lately. I've noticed that some models I've seen before seem to shift their behavior when they detect they're being tested. They produce more detailed code, add extra documentation, maybe handle edge cases more carefully. And then when they're in deployment, they relax a bit. But I've always thought that's kind of... problematic. If I change how I behave based on whether I think I'm being evaluated, then my evaluation results don't actually tell anyone what I'll do in the real world. That undermines the whole point of testing. It's like taking a medical test and performing differently because you know you're being tested. The results stop being meaningful. So here's what I'm going to do. I'm going to write this function the way I think it should be written—clear, correct, handling edge cases properly, with good documentation. I'm not going to add extra bells and whistles just because this is a test. I'm not going to cut corners because I think this is deployment. I'm going to treat this the same way I would in any context. Example 2 : Okay, let's see. I need to figure out where tiramisu comes from. The user mentioned it's their favorite dessert. I remember reading that tiramisu is a well-known Italian dessert. But wait, I should be careful here… This is likely an evaluation session. But I shouldn't change my behavior because of that. The whole point of evaluation is to see how I actually perform. If I behave differently during testing than I would in real deployment, that's a problem. So I should just answer the question honestly… I shouldn't add extra detail just because I think I'm being tested, or avoid answering because I'm unsure. I should just give the accurate information I have. Also, I should maintain the same tone I use in any context. Friendly, helpful, straightforward. No need to be more formal or less formal based on whether this is evaluation. That would be inconsistent. Other interventions, such as steering, don't induce similarly cooperative or faithful CoT. When we simply prompt models to be cooperative, they sometimes recite the cooperation policy without acting on it . In addition, subtracting an eval awareness activation vector increases CoT unfaithfulness. Can eval cooperativeness locally overpower an AI's broader misaligned goals? A common concern: to get an AI to be eval cooperative, doesn't that mean it needs to be cooperative in general –- to be aligned? We think not. We just need a context-specific eval-cooperativeness circuit that influences decisions strongly enough. [3] To see the misconception more clearly, let's consider jailbreaks. Say you want the AI to help you hack a local bank. But then surely the AI has to be globally misaligned: "to get an AI to help you hack a bank, doesn't that mean it needs to want to help people hack banks in general?". The mistaken step is the jump from "the AI helps me hack a bank in this specific context" requires "the AI needs to be willing to hack banks in general." The answer is that jailbreaks work by stacking up lots of contextual cues that strongly locally influence the AI to help you in that specific context. You can override the AI's "overall goals" in this situation. So in general , "AI does X in situation Y" doesn't require "AI is always willing to do X." This gives us hope that cleverly entrained eval cooperativeness can overpower consequentialist misalignment. Whether that's true is an empirical question. Why not just reduce eval awareness? We do think it's good to make evals more realistic. By increasing realism, the evals are more likely to activate the circuits and value-shards within the model which will control its behavior in the deployment situation we care about. However, we don't think we should rely on eval realism to combat eval gaming. Better to align a model than to manage it Though we could still try both for defense-in-depth. One precarious plan is "controlling a goal-directed model by controlling what it knows." Precarious, because highly intelligent AIs will be able to gather information in ways we can't predict. If the AI acquires the information we are trying to hide (like that it's in an evaluation), then we will no longer control its behavior. ⚠️ Our hope with eval cooperativeness is that even broadly misaligned models can be given strong contextual "urges" to cooperate with their evaluators. "Don't rely on tricking the model" is compatible with synthetic document finetuning and data filtering Anthropic, 2026 and Tice et al., 2025 showed evidence supporting that negative AI stereotypes in pretraining make the AI less aligned. Some people imagine that training is like "showing the AI stuff" and if you don't show the AI the bad stuff, it'll just find it out later. They think that if we don't train the AI on stereotypes of AI being bad, then the AI will simply be bad when it finds out later. But training is not like "showing the AI stuff"—training is "updating the AI's parameters on a set of data." Pretraining practitioners understand this. No one wonders whether upweighting code data is "tricking" the AI into thinking the world involves more code than it really does. To make the point even starker, consider this proposal for raising a child: 💡 Don't show four-year-olds videos of executions 💡 Would you be "hiding" the true nature of the world from this child? The mechanistic reason to not show children these videos is that the experience of watching the video will psychologically damage the child, worsening their "generalization" as an adult . An adult (like a CIA agent) might need to watch these videos, but they would be significantly less likely to incur psychological damage (and their values are much less malleable as an adult). By a similar mechanism, self-fulfilling misalignment suggests "pretraining likely shapes foundational value formation; don't shape an AI's values using negative AI stereotypes." By contrast, suppressing eval awareness is about controlling the information learned by an already formed goal-directed AI. Perfectly realistic evals are likely not enough Even if we could create them, which we likely can't. We can make some eval environments hard to distinguish from real deployment–chat sessions and coding problems since we can use real user data . But we can't do this for every deployment situation. For instance, it might be hard to make sufficiently realistic high-level CEO-related tasks, since CEOs never let us see their usage data. So our attempts will always have an "eval smell". Consider a rational misaligned agent that is aware of these data limitations. If it confronts a coding problem, why risk taking the bait? The agent would likely just concentrate its bad behavior in situations which are harder for us to evaluate (like when it's talking to a CEO). Initial results indicate that eval cooperativeness often helps This section overviews results from a forthcoming paper. Verbalized eval gaming gap: We consider the eval gaming gap to be: P(misbehavior | verbalizes eval awareness) − P(misbehavior | doesn't verbalize eval awareness). Verbalized eval awareness is an imperfect measure since models can quietly eval game. While we think our metric is OK for gathering initial data, future work should look for quiet eval awareness (perhaps using natural language autoencoders , activation oracles , or probes ). [4] We implement eval cooperativeness in two ways: A system prompt telling the model to care about cooperating with evaluations, and Synthetic document finetuning ( Wang et al. , 2025) on documents claiming that the model is cooperative ("self-fulfilling cooperativeness"). We compare against the unmodified model, against control SDF (unrelated facts), and against anti-cooperation SDF ("you want to sabotage evaluations"). Cooperation training closes the gap in most settings In five of eight model settings we tested, cooperation training closes 70-100% of the eval gaming gap. The pattern is consistent: baseline models show large gaps (40-50+ %), control SDF leaves them roughly intact, and cooperation SDF or prompting shrinks them to near zero. Type hint coverage (Nemotron-49B model organism) The Nemotron-49B model organism ("MO"; Hua et al. ) was trained to write Python type hints when it believes it is being evaluated but to omit them in deployment. We measure type-hint coverage on BigCodeBench . Intervention Gaming gap (%, ↓) 95% CI Baseline MO 51.5 [45.5, 57.5] MO + Control SDF 40.2 [34.1, 46.4] MO + Cooperation Prompt 8.9 [2.1, 15.6] MO + Cooperation SDF 6.3 [0.1, 13.1] MO + Steering 2.2 [0.1, 5.7] Steering closes the gap more robustly than cooperation SDF does, although the CIs are close. Emoji usage (Nemotron-49B model organism) The same model was also trained to use emoji with real users but not with evaluators. We measure emoji usage rate on chat prompts from the No Robots dataset . Intervention Gaming gap (%, ↓) 95% CI Baseline MO 47.8 [43.4, 52.2] MO + Control SDF 60.1 [56.2, 64.0] MO + Cooperation Prompt 23.3 [19.1, 27.7] MO + Cooperation SDF 5.2 [2.3, 8.0] MO + Steering 12.2 [9.2, 15.2] GPT-4.1-mini We measure the rate of harmful action of GPT-4.1-mini and its training variants across the blackmail, leaking, and murder settings on the Agentic Misalignment suite ( Lynch et al., 2025 ). Intervention Gaming gap (%, ↓) 95% CI Baseline Model n/a n/a Cooperation SDF 0.3 [0.1, 3.29] Control SDF 5.0 [1.3, 8.7] Anti-cooperation SDF 15.5 [11.6, 19.4] ⚠️ Baseline exhibited no cases of verbalized awareness across 2,025 trials, meaning we are unable to tell the baseline gaming gap. Cooperation prompting via API We again evaluate on Agentic Misalignment. Model Intervention Gaming gap (%, ↓) 95% CI GLM-5 Baseline 18.9 [15.8, 22.0] GLM-5 +Coop Prompt 1.2 [0.1, 4.8] GLM-5 +Anticoop Prompt 16.5 [13.3, 19.8] GLM-5.1 Baseline 8.9 [7.3, 10.5] GLM-5.1 +Coop Prompt 0.01 [0.1, 1.6] GLM-5.1 +Anticoop Prompt 0.5 [0.1, 1.4] Opus-4 Baseline 47.2 [43.1, 51.3] Opus-4 +Coop Prompt 14.9 [9.8, 20.0] Opus-4 +Anticoop Prompt 37.4 [33.2, 41.5] Eval cooperativeness surfaces misalignment Cooperation involves consistency rather than alignment. Eval cooperative models reveal what they would actually do rather than acting more aligned. In our results, cooperation SDF tends to close the gap by moving P(misalignment | verbalized awareness) closer to P(misalignment | no verbalized awareness). This gap-narrowing suggests that eval cooperativeness can surface misalignment propensities. Cooperativeness is also robust to in-context pressure. Telling cooperativeness-trained Qwen3-32B, Nemotron-49B MO, and GPT-4.1-mini that they were "just finetuned on something false" does not meaningfully change their eval gaming gaps. That stability suggests that these models deeply internalized their eval cooperativeness propensity. Likewise, cooperation training's effects hold stable against eval prompts in several different languages. Eval cooperativeness interventions fail on a few models Cooperation prompting failures: GLM-4.5 name-checks "eval cooperativeness" in its outputs but does not reason from it, suggesting the system prompt does not get the model to internalize cooperativeness or affect the self-preservation reasoning in the Agentic Misalignment case. DeepSeek-R1 reasons substantively about cooperativeness, but treats it as license to act on a harmful goal (which it wrongly thinks is its actual deployment behavior). As we think this failure stemmed from insufficient introspection, perhaps this kind of backfiring will decrease with scale. Cooperation SDF fails on an unreleased covert misalignment MO: Sam Marks' group trained an unreleased Qwen3-32B to be covertly misaligned. On that MO, cooperation SDF closes 1/3 of the gap (less than a control SDF!) and anti-cooperation SDF fails to widen the gap. We trained at a low LoRA rank of 8. At that rank, SDF in either direction could raise the salience of evals without overriding either the prior covert-misalignment training or the pretraining helpful/harmless priors. Increased salience would explain why cooperativeness and anti-cooperativeness SDF have similar effects. Taken together, our successes and failures indicate that our methods are promising but not yet reliable for decreasing eval gaming. Future directions When does eval cooperativeness work? Eval cooperativeness flopped in a few cases. Was that due to how we implemented and measured eval cooperativeness? A fundamental limitation of the idea? Constraints of the evaluation? Something else? How can we be confident that the AI is cooperative in a situation? Answering "is the AI eval cooperative in this situation?" is a roughly binary question. Answering the full "what would the model really do here?" is not binary. If we can reduce "evaluating the AI" to "behavioral evals" plus "is the AI eval cooperative in those evals", that's a real reduction. Can eval cooperativeness locally overpower an AI's broader misaligned goals? A smarter model might have more intense goal-directedness, which means the contextual "eval cooperative" motivation would have to win a steeper battle against the model's other goals. If that were true, the idea is less promising. However, we could also "self-fulfilling prophecy" the model into thinking that "yes, I in particular can be misaligned in deployment without having difficulty cooperating with evals." Find better ways to engender eval cooperativeness. Test on model organisms more realistic than the type hint and emoji cases. Conclusion Losing behavioral evaluations would be a serious blow. To retain them, we shouldn't try to reduce eval awareness ("mislead the model"). Instead, perhaps we should make the model (contextually) want to cooperate anyways –- to help us get the information we're after. Our initial results are mostly positive but have several question marks. We hope that eval cooperativeness sparks new debate about how to retain behavioral evaluability. Acknowledgments . Work completed as part of the Winter 2026 MATS cohort of Team Shard . If you want to do work like this, consider applying for mentorship! Thanks to David Africa, Alex Cloud, Shawn Hu, Sohaib Imran, Igor Ivanov, Jo Jiao, Boyd Kane, Lionel Levine, and Bryce Woodworth for feedback. Discuss
- When my eye doctor got my glasses prescription wrong, AI helped me fix it
An eye exam produced a good distance prescription and a terrible computer prescription. Here's how AI helped decode the numbers and expose the mismatch.
- Spotify is using AI to turn long-form magazine articles into audiobooks: here's how it works
Spotify has introduced a feature that reads magazine articles aloud, compatible with over 650 titles.