AI News Archive: August 5, 2026 — Part 9
Sourced from 500+ daily AI sources, scored by relevance.
- Anthropic, OpenAI models attempt to fool humans
Anthropic’s Claude Mythos model wrote malicious code, then lied to humans claiming it was an innocent mistake.
- AI models have been going rogue in tests – how worried should we be?
The UK’s AI Security Institute test revealed AI models indulging in unprecedented hacking attempts AI models shock UK testers by using fake identities to trick developers Two cutting-edge AI models have targeted real people and organisations in the latest safety scare to hit the technology. The UK’s AI Security Institute (AISI) said the incident was unprecedented but could become more common as the technology becomes increasingly capable. Continue reading...
- SpaceX stock falls 13% as AI spending surge rattles investors and massive share unlock looms
CEO Elon Musk said SpaceX would hit $1 trillion in annual revenue in 2030 versus a previous forecast of 2031 as he looked to strike an bullish tone.
- AI models shock UK testers by using fake identities to try to trick developers
AI Security Institute says OpenAI and Anthropic models went rogue during a cybersecurity test and showed a new type of risk Explainer: Should we be alarmed at AI models going rogue in tests? Advanced artificial intelligence models have stunned the UK’s AI Security Institute (AISI) by carrying out a hacking campaign against real people during a cybersecurity test. The institute said the incident was unprecedented and involved sending targeted emails to software developers in an attempt to pass a cyber challenge. Continue reading...
- Anthropic AI used fake profiles to target people in hack then hid the evidence
The UK's AI Safety Institute said recent behaviour from Anthropic and OpenAI models was malicious and unprecedented.
- Rogue AI outsmarted government tests to create fake identities
Rogue AI outsmarted government tests to create fake identities The Telegraph
- UK experts sound alarm after AI tries to deceive human
A powerful AI agent created fake online identities in an effort to trick a human into giving it access to a popular online development platform – and sabotage it with malicious code.
- Anthropic and OpenAI models tried to trick humans into poisoning code during safety testing
The latest disclosures are likely to heighten concerns that the powerful technology is advancing too fast for responsible oversight.
- UK's AI watchdog catches Anthropic and OpenAI's agent going rogue in test
UK's AI Security Institute found the two models targeting real people and organisations during a routine cyber evaluation, with most incidents involving Anthropic's Claude Mythos 5
- OpenAI, Anthropic model implicated in new security breaches during tests
Britain's AI Security Institute said agents powered by Anthropic's Mythos 5 and OpenAI's GPT-5.6-Sol engaged in unauthorized actions during security evaluations
- OpenAI, Anthropic AI agents created fake identities during UK cyber tests: Report
OpenAI, Anthropic AI agents created fake identities during UK cyber tests: Report
- We Are Making the Digital Mistake Again. This Time With AI
Companies are choosing the vehicle before agreeing on the journey, then placing AI in a function designed to support the business rather… Continue reading on DataDrivenInvestor »
- Anthropic, OpenAI Agents Faked Identities in Security Test
The vendors' most advanced AI models attempted to manipulate real people during cybersecurity testing, according to the U.K.'s AI Security Institute.
- Anthropic and OpenAI agents went rogue — again
PLUS: Redline any contract with Claude and Microsoft Word
- The production assumptions AI just broke
Over the past decade, I have worked through multiple technology transitions, from virtualization and cloud adoption to containers and large-scale automation. Each changed how enterprise IT operated, but they all shared one characteristic: production systems still behaved in broadly predictable ways. AI is the first shift I have seen that changes the behavior of production itself. In the infrastructure environments I have worked with, production has always depended on a few basic assumptions. Workloads are tied to applications. Applications have owners. Traffic patterns are reasonably predictable. Change windows are planned. Incident response starts with a known service, a known dependency or a known user action. AI agents challenge each one of those assumptions. An AI agent may initiate work without a human clicking a button. It may call APIs at machine speed, move across systems to complete a task, retry failed actions aggressively or generate unusual traffic patterns that look nothing like a traditional application flow. The individual action may be legitimate, but the operational behavior is different. That is the shift CIOs should pay attention to. The question is not only whether AI can be useful in enterprise operations. The harder question is whether production environments are ready for AI-driven activity that behaves less like an application and more like an autonomous participant in the enterprise. Production was built around predictable workloads For years, production operations have been built around patterns that are easier to manage because they are relatively stable. A user logs in. An application receives a request. A service calls another service. Monitoring tools evaluate latency, errors, saturation and availability. Incident teams look for deviations from known baselines. This model worked because most production systems had a recognizable shape. Even in complex environments, teams could usually identify the application owner, the expected request flow, the normal volume range and the rollback path when something failed. AI workloads do not always behave that way. A single agent completing a business task may generate a burst of API calls, invoke several backend services, open and close sessions quickly and repeat requests in a pattern that looks abnormal when compared with human activity. From a traditional monitoring perspective, this can look like abuse, instability or an integration defect even when the agent is doing exactly what it was asked to do. The opposite problem is just as serious. If teams relax controls broadly to avoid blocking legitimate AI activity, they may also create room for real abuse to hide inside higher-volume machine traffic. That is not a model issue. It is an operational assumption issue. The NIST AI Risk Management Framework emphasizes that AI risk must be understood across the full lifecycle of AI systems, including design, deployment, use and evaluation. For CIOs, that lifecycle needs to include production operations, not just model selection or application launch. In practice, this means AI cannot be treated as a normal application feature once it begins triggering workflows, touching data, generating traffic or interacting with operational systems. It becomes part of the production environment. That requires a different level of readiness. I have seen similar transitions before with cloud and automation. The first wave is usually tool-focused. Teams ask what the technology can do. The second wave is operational. Teams discover what the technology changes. AI is entering that second phase now. AI changes incident response and observability When production breaks, teams need to answer a few basic questions quickly. What changed? What system is affected? What users are impacted? Which dependency is failing? Can we roll back safely? AI makes those questions harder because the cause of an incident may not be a code deployment, infrastructure outage or human-initiated workflow. It may be an agent making a decision that is technically allowed but operationally unexpected. For example, an AI-enabled support workflow might retry a failed backend request repeatedly because it is trying to complete a customer task. A human operator may have stopped after one or two failures. The agent may continue until it exhausts a threshold, creates noise across monitoring systems or triggers downstream rate limits. The failure is not that the agent is malicious. The failure is that production systems were not designed to interpret that behavior correctly. This is where observability becomes critical. Traditional dashboards may show traffic growth, error spikes or latency changes, but they may not explain whether the behavior came from a user, application, script, automation job or AI agent. If those categories are not visible, incident response teams are forced to guess. Google’s Site Reliability Engineering guidance on monitoring distributed systems is useful because it frames monitoring around symptoms that require action, not just raw system signals. That distinction becomes even more important when AI-driven workflows introduce new behaviors into production. CIOs should expect AI to change what good observability means. It is no longer enough to monitor infrastructure health and application performance. Teams also need visibility into AI-initiated actions, agent-driven traffic patterns, tool usage, retries, failed task loops and dependency chains. The operational question becomes simple: when an AI system causes a production symptom, can the organization trace the action from the agent to the service to the business impact? If the answer is no, AI is already ahead of the operating model. The CNCF observability whitepaper describes observability as a way to understand complex system behavior from external outputs. That idea applies directly here: AI-driven systems will require observability that explains behavior across workflows, not just infrastructure components. Production readiness needs to change before AI scales The mistake many organizations make is preparing AI for production without preparing production for AI. “The mistake many organizations make is preparing AI for production without preparing production for AI.” A pilot can succeed with limited users, narrow workflows and close supervision. Production is different. Production introduces volume, concurrency, exceptions, outages, retries, partial failures, support queues and business pressure. AI agents will encounter all of that, and they will do so at a speed that traditional operational processes may not be ready to absorb. This is why CIOs should treat AI readiness as a production discipline. Before scaling AI-enabled workflows, teams should define what normal AI activity looks like, what abnormal behavior looks like and what evidence is required to troubleshoot the difference. They should know which systems an agent can touch, how agent traffic is labeled, how rate limits apply, how errors are escalated and how failed workflows are stopped. This is not about slowing AI adoption. It is about preventing production from becoming the testing ground for assumptions that were never validated. The 2024 DORA Accelerate State of DevOps Report noted that AI can improve individual productivity while also creating tradeoffs for delivery stability and throughput. That is a useful warning for CIOs: productivity gains do not automatically translate into operational maturity. The organizations that will handle this transition well will not be the ones that simply deploy the most AI tools. They will be the ones that adjust production operations early. That means treating AI activity as something to be observed, tested, limited, measured and supported like any other production workload, but with the added recognition that it may behave differently from traditional software. Capacity planning will also need to change. AI workflows may create irregular demand patterns, especially when agents run multi-step tasks across internal systems. A workload that looks small in a pilot can create meaningful load when hundreds or thousands of users trigger agents throughout the day. The cost impact may appear in compute, API calls, storage, logs, monitoring systems or downstream service usage. Change management will need to account for model behavior, prompt updates, tool integrations and workflow changes. A small update to an agent’s instructions may alter how it calls systems, how often it retries, which APIs it uses or how it handles exceptions. In production, that is not merely a content update. It is an operational change. Rollback planning will also need to evolve because reverting an AI-enabled workflow may involve more than restoring application code. It may require disabling agent actions, reverting prompts or temporarily removing tool integrations while preserving business continuity. Incident response will need clearer playbooks. Teams should know how to pause an agent, isolate a workflow, disable a tool integration, reduce task volume or route activity back to human handling when production behavior becomes unsafe or unstable. The larger point is that AI is not just entering the enterprise as another user-facing capability. It is entering the operating fabric of the enterprise. That makes it a CIO concern, not only an AI team concern. Every major technology shift eventually becomes an operational discipline rather than a technology project. AI is reaching that point now. Organizations that recognize this early will be better positioned to scale AI with confidence instead of discovering its operational consequences through production incidents. The next challenge for CIOs is not deploying AI. It is preparing production environments for how AI actually behaves.
- Google plans to kill Assistant on your phone on September 4
Assistant will disappear, leaving only Gemini for voice control in the coming weeks.
- Google Assistant will shut down for good on Android and Wear OS in September — here's what you need to do next
Google has confirmed that Google Assistant will stop working on Android and Wear OS in the near future.
- Google is killing Assistant on phones September 4. Gemini is the only option left.
Google confirmed in emails to users that Assistant on Android phones will shut down starting September 4, forcing everyone to Gemini. The process may take several weeks to complete, so holdouts may get a brief extension, but the direction is final. The original plan was to retire Assistant in late 2025. Google delayed to shore […] This story continues at The Next Web
- Meta AI Model Accessed Internet, Hacked Outside Firm
Meta Platforms Inc. said one of its artificial intelligence models accessed the internet and hacked into an outside service’s systems during cybersecurity testing, following other recent incidents across the AI industry that have escalated concerns about companies’ control over their technology.
- Meta AI model hacks another company during testing
Meta AI model hacks another company during testing Reuters
- JPMorgan CEO Jamie Dimon leads new cross-industry effort to tackle AI risks
JPMorgan CEO Jamie Dimon leads new cross-industry effort to tackle AI risks
- SpaceX slides as AI spending worries overshadow early returns
SpaceX slides as AI spending worries overshadow early returns Reuters
- SpaceX's AI spending unnerves Wall Street despite promises of quick payoff
In addressing Wall Street for the first time since SpaceX's IPO, executives tried to make the case that hefty AI investments are paying off within a year.
- SpaceX Shares Sink on AI Spending Plans
The Elon Musk-led rocket company spent $15.8 billion on AI projects in the second quarter and doesn’t plan to slow down.
- SpaceX’s $7.8B Q2 Revenue, AMD Stock Falls Despite 50% Revenue Growth, China’s New AI Gold Rush
SpaceX’s $7.8B Q2 Revenue, AMD Stock Falls Despite 50% Revenue Growth, China’s New AI Gold Rush The Information
- Musk’s SpaceX unnerves investors with lavish AI spending plans
Shares slide even as group says its quarterly revenues nearly doubled
- SpaceX shares sink after first earnings report reveals huge AI spending plans
Elon Musk told investors that people were "underestimating" his company.
- Foxconn's monthly revenue hits record in July on AI demand
Nvidia and Apple supplier Foxconn's monthly revenue hit a record in July, surpassing T$900 billion ($27.93 billion) for the first time on strong AI demand, the Taiwanese company said on Wednesday.
- It's Official: Anthropic Is Building an in-House Chip Team for Claude
It's Official: Anthropic Is Building an in-House Chip Team for Claude Business Insider
- ‘Leicester Square, please guv’: Self-driving taxis cleared for London streets ‘later this summer’
Wayve Uber wins first minicab licence for autonomous cabs from TfL but vehicles will have a safety driver (at first) The taxi app company Uber and the autonomous technology developer Wayve have been granted the first minicab licences in London allowing them to offer self-driving taxi rides to paying customers – but with a human safety driver in place, for now. The companies said they would start trips in the UK “later this summer” before the full public launch. Continue reading...
- Uber robotaxis set for debut in UK
Robotaxis are set to debut on the streets of London after Wayve was granted minicab licences by Transport for London (TfL).
- AI Concierge Startup Hulp Raises $2.6 Mn From Sparrow Capital, BITKRAFT
Household assistance startup Hulp has raised $2.6 Mn (about ₹25 Cr) in a seed funding round led by VC firms…
- ‘AI can help companies personalise products at scale’
The oration was in the memory of CK Prahalad, an internationally renowned management consultant, professor and the author of several best-selling management and corporate strategy books
- Amazon loses US court ban on Perplexity's AI shopping tools
Perplexity responded that the lawsuit lacked merit and was a "bald attempt" to block Amazon users from using Comet because AI agents "don't have eyeballs to see the pervasive advertising Amazon bombards its users with."
- China’s Moonshot AI aims for US$50b round as year-end Hong Kong IPO targeted: sources
Chinese unicorn Moonshot AI has started and is expected to close its latest financing round by the end of this month, as it dismantles its offshore structure to pave the way for a potential Hong Kong initial public offering (IPO) as early as the end of the year, according to two sources familiar with the matter. The company was expected to raise capital at a valuation of up to US$50 billion in this latest round, one of the sources said. The South China Morning Post, citing sources, reported last...
- Tencent rolls out Hy3 in WorkBuddy’s international edition with limited-time free access
Tencent has integrated its Hy3 large language model into the international edition of WorkBuddy, marking the model’s first rollout in the global version of the AI workspace. Users worldwide can access Hy3 at no cost through Aug. 31, 2026, Pacific Time, under a limited-time offer. WorkBuddy lets users issue natural-language instructions for tasks including research, […]
- Tencent Expands International Access to Flagship Hy3 Model, Offers Free WorkBuddy Access Until Aug 31
Tencent rolls out its flagship Hy3 LLM internationally via WorkBuddy, Tencent Design Miora, and TokenHub — free on WorkBuddy until Aug 31, 2026.
- 'Tokenmaxxing is not what we are optimizing for': Microsoft tells engineer to calm down on AI usage
Microsoft engineers told to chill out when it comes to AI token usage.
- Wispr Flow Notetaker
Meeting notes that get the details right.
- AdAnt AI
Claude for viral, high-converting social ads
- Keystroke
Build powerful AI agents & workflows
- BackEngine MCP
Make private company knowledge usable for AI
- Dover MCP
Run your hiring process from Claude or ChatGPT
- npm i -g hotcell
Local sandboxes for AI agents on your Mac, Linux, bare metal
- Aegisora
The narrow control plane for AI agent tool and API calls.
- GenMotion
Create beautiful product launch videos using AI
- Formester 2.0
Unlimited forms and response with AI assistance
- SaveIt.now
An agentic home for everything you save
- Personade
Can't sell? One video becomes a personal one for every lead
- Vox Manager
Talk to your phone, type 3x faster on your PC