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
Pope Leo XIV’s first encyclical, Magnifica Humanitas, makes artificial intelligence a matter of human dignity, public policy and democratic control. The document argues that AI should serve human flourishing rather than quietly persuading people to treat human limits as defects and machine intelligence as a replacement for human…
Read the report & discussion ↗Watch Desk
Google Research has published a retrieval system that shifts the expensive part of complex search from the moment you ask a question to an earlier training stage. Its Retrieve-for-Train framework is designed to produce diverse, complementary search results in a single pass, rather than making a large language…
Meta AI Watch
More than 70 people have sued Meta over its AI-enabled smart glasses, alleging that the devices captured intimate footage and sent material to workers involved in training Meta’s artificial-intelligence systems. The allegations matter because they challenge the gap between a recording light, user consent and what an AI…
Watch Desk
Anthropic says Claude now performs 26% of its model research and development, while OpenAI and xAI are also pursuing increasingly automated research. The important distinction is that AI is helping build better AI, but fully autonomous recursive self-improvement remains a goal, not an achieved capability.
Geoffrey Hinton Watch
Geoffrey Hinton has backed a proposed rulebook for independent AI evaluators, calling for access inside companies such as OpenAI and Anthropic, protection from interference and the freedom to publish uncomfortable findings. The important point is that this is not another general plea for “more safety”. It is a…
OpenAI Watch
OpenAI and Microsoft executives made private and courtroom statements about AI systems replacing journalism, according to newly public filings reported by Reuters. The remarks give publishers a sharper argument in a case that could help decide whether training AI on millions of news articles counts as fair use.
OpenClaw Watch
OpenClaw has launched a public security hub that gathers its advisories, vulnerability-reporting process and ongoing hardening work in one place. The open-source AI project says it has published 722 fixes since January, including fixes for 14 reports classified as critical vulnerabilities.
Yoshua Bengio Watch
Yoshua Bengio says governments may be approaching a decisive turning point on AI safety, while his non-profit LawZero receives up to C$300m in announced Canadian and German government funding to develop a guardrail for autonomous systems. The important point is not another warning about AI in the abstract.…
Watch Desk
Cybersecurity companies are taking sharply different routes into AI, from purpose-built defender models to frontier systems wired into existing security platforms. The practical problem is that buyers are being handed metrics measured in different units, making the next big security purchase rather harder to compare than the marketing…
Watch Desk
Former US antitrust chief Jonathan Kanter says frontier AI labs do not need an antitrust exemption to cooperate on safety. In a Decoder interview, he argues that companies can share threat intelligence lawfully, while any pact to slow competition could trigger the very rules they want waived.
Ilya Sutskever Watch
Ilya Sutskever’s Safe Superintelligence has reportedly raised roughly $8bn while shipping no public product, paper or demo. That makes the lab’s silence unusually expensive, and turns its secrecy into the central AI business bet.
Watch Desk
The campaign to slow frontier AI has collided with the industry’s favourite counterargument: China. A Guardian analysis shows how Dario Amodei, Donald Trump and a wider circle of technology leaders invoke strategic competition when deciding how much restraint the United States can afford. That matters because a safety…
Watch Desk
TypeSafe AI's Jev is being pitched as a different kind of AI model: not a chatbot, not a reasoning engine and not a replacement for your main LLM. Theo Browne describes it as a fast classifier for structured decisions such as routing, ranking, tagging and escalation, where the…
Watch Desk
The director of a Swiss AI research institute has entered the doom debate with a firm shrug at the extinction scenarios. Andrea-Emilio Rizzoli, who leads the Dalle Molle Institute for Artificial Intelligence (IDSIA USI-SUPSI), says the warnings driving this month's headlines are highly speculative, and that the risks…
Meta AI Watch
Meta's biggest event of the year is five days away, and the most intriguing product in the expected lineup has no camera at all. Connect 2026 opens with Mark Zuckerberg's keynote at 4PM PT on 23 September, midnight for UK readers, with developer sessions following on the 24th.…
Anthropic Watch
Anthropic's threat intelligence team has laid out its fullest account yet of attempts to turn Claude into a weapon, and the case studies run from bioweapons grant drafts to a kamikaze drone swarm. By the company's own account, it disrupted every operation described.
Anthropic Watch
Anthropic is letting outsiders through the lab door. The company announced on 18 September a partnership with Accenture to embed independent evaluators inside Anthropic with access comparable to an employee's, and each side expects to invest at least $1 billion in building that capacity over the next five…
Watch Desk
California is drawing up plans to switch off AI. Governor Gavin Newsom has issued an executive order asking state agencies to recommend stronger independent reviews of the technology and to consider a shut-off function should it get out of control, while a convened panel of experts has two…
Zhipu AI / Z.ai GLM Watch
Zhipu's chief scientist Tang Jie says the lab has reached a "minimal closed loop" in recursive self-improvement: an agent driven by its GLM-5.3 model helped build and optimise the inference service that serves GLM-5.3-Flash. If the numbers hold, a model is now doing engineering work on the system…
Watch Desk
Johns Hopkins University convened an evening on AI governance in Washington on Monday, and the exchange did something the week's op-ed war has not: it put an OpenAI executive, two prominent governance scholars and one very large demand on the same stage. The demand is that AI developers…
