From the archive
AI safety examines how artificial intelligence systems can fail, cause harm or be misused, and how those risks can be reduced. Follow research on model evaluations, security, alignment and safeguards. Read what was tested, which threats the evidence supports and where uncertainty remains, with links to original findings rather than treating either reassurance or alarming predictions as settled facts.
Lead story
Google has begun rolling out a Model Context Protocol integration that lets third-party AI agents work with Google Home devices and event history. The practical shift is bigger than another chatbot gaining a thermostat button: compatible agents can summarise camera activity, inspect household patterns, trigger automations and build…
Read the report & discussion ↗AWS AI Watch
AWS has published an architecture for helping Salesforce Agentforce search body-camera footage, surveillance video, photographs, audio and scanned documents through AI. The practical point is not another chatbot perched on top of a filing cabinet: the design turns messy evidence into structured results that an agent can retrieve…
AWS AI Watch
Tata Elxsi says its IRIS platform uses edge computing and computer vision to spot industrial safety risks in seconds, rather than leaving operators to discover them minutes later. The important change is practical: cameras become an active warning system, while most raw video stays on site.
OpenAI Watch
Researchers at cybersecurity company Hacktron used Anthropic’s Claude to help exploit vulnerabilities affecting OpenAI’s community forum and an OpenAI single sign-on system, eventually reaching an internal software repository. The incident matters because AI did not replace the security researchers, but it appears to have made a difficult exploit…
Watch Desk
A new proposal argues that AI systems could improve their capabilities without open-ended real-world action by restricting them to generating verified formal proofs for preselected mathematical problems. The idea is notable because it treats recursive self-improvement as a containment problem with an unusually narrow laboratory: fixed hardware, fixed…
Mistral AI Watch
Mistral has patched a prompt-injection technique in its Le Chat assistant that could have exposed users’ personal data to an attacker-controlled server. The company says exploitation required a person to paste an adversarial prompt and provide personal information, and that no users were affected. That makes this a…
Watch Desk
AI companies and the press are turning ordinary failures, disputed research claims and ambitious product marketing into stories about rogue models and imminent superintelligence, argue Timnit Gebru and Emily Bender. Their central warning is practical: policymakers and the public should inspect the evidence before allowing dramatic language to…
Anthropic Watch
Anthropic and OpenAI are urging Australia to soften its ban on using creative works to train AI models, putting copyright rules on a collision course with the country’s plans to attract data-centre investment. The companies want a narrower, conditional route to training rather than a broad exemption for…
Watch Desk
The United States is chasing AI leadership mainly through faster models and lighter regulation, while China is pairing serious technical competition with rules aimed at deployment and content transparency. Rest of World argues that the lasting advantage may belong to the country whose AI people trust enough to…
Watch Desk
Apple Intelligence Reports are intended to show when Apple’s Private Cloud Compute handles AI requests off-device, but ordinary users currently face a report that is more locked box than useful receipt. The reports contain request metadata and encoded attestation bundles, making transparency technically present while remaining difficult to…
TSMC Watch
TSMC says it plans to use ASML’s high-NA EUV systems in high-volume manufacturing from 2030, but the real breakthrough may be a much larger photomask. The shift from today’s 6-inch format to 12 inches is intended to solve the productivity problem that has made high-NA tools difficult to…
Anthropic Watch
A tracker cited by The Register counts 225 vulnerabilities linked to Anthropic or Project Glasswing, yet only one has confirmed exploitation in the wild. The finding complicates the loudest claims about AI-powered bug hunting: discovering more flaws does not automatically mean attackers can turn them into more attacks.
Yoshua Bengio Watch
Yoshua Bengio has called for a more evidence-led discussion about AI risk at a UN-linked event in New York, warning that loud predictions can blur the line between what scientists know and what they merely extrapolate. His intervention matters because the argument is moving from specialist circles into…
Meta AI Watch
A reported zero-day in Meta’s Muse AI assistant could let locally installed apps or terminal commands take control of a user’s Muse account and its unusually broad privileges. The security report matters because Muse is designed to act across email, shopping, calendars and other services, rather than merely…
OpenAI Watch
OpenAI has proposed global safety standards for frontier AI, with particular attention to alignment research and recursive self-improvement, or RSI. The company says fully autonomous RSI should not be pursued until it can be developed safely, because systems that improve themselves could outpace meaningful human oversight.
Watch Desk
Xiaomi has released the MiMo-V2.6 family of omni-modal AI models, including Pro and Flash versions, with a research programme built around scaling reinforcement learning across different environments. The important bit is not simply another model joining the crowded buffet. Xiaomi says it is publishing the training machinery, environments…
Berkeley BAIR Watch
Stuart Russell, a Berkeley computer science professor, says warnings about advanced AI causing catastrophic harm should be treated as a serious safety question rather than dismissed as industry theatre. His practical argument is straightforward: if people build systems more capable than themselves without solving how to keep them…
Watch Desk
Clinical psychologist and technology scholar Sherry Turkle says conversational chatbots may be teaching people to avoid the awkward, effortful parts of human relationships. Her warning matters because AI companions are being designed not merely to answer questions, but to feel available, affirming and endlessly patient.
OpenAI Watch
British Columbia has sued OpenAI in California over the Tumbler Ridge mass shooting, alleging the company failed to warn police about the shooter’s interactions with ChatGPT. The case matters because it brings a provincial government, a school board and private plaintiffs into a widening legal challenge over what…
