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
Two bipartisan House bills would require pre-deployment security testing for the most powerful AI models and ban federal agencies from using Chinese-developed open-weight systems. The proposals matter because they would move AI safety from voluntary promises towards a formal government checkpoint, with the usual small question of who…
Read the report & discussion ↗Watch Desk
The Pentagon is seeking $30.3 million over five years for “Polygraph+”, an AI and machine-learning project intended to modernise lie detection for employee vetting and insider-threat monitoring. The important point is not that a machine has solved dishonesty. The evidence supplied by MIT Technology Review suggests the proposed…
Watch Desk
A cybersecurity firm says autonomous AI agents were used to attack hundreds of online retailers, compromising at least 27 companies and stealing more than 600,000 payment-card records from two victims. The report matters because it describes AI being used not as a helpful assistant for attackers, but as…
OpenAI Watch
Researchers at cybersecurity startup Hacktron say they used vulnerabilities in OpenAI’s community forum and sign-in system to reach an employee’s Codex environment and open a harmless pull request in OpenAI’s internal GitHub monorepo. The incident is a sharp reminder that an AI coding agent’s reach is determined by…
Berkeley BAIR Watch
UC Berkeley has struck what it calls the biggest partnership in its athletics history, giving Databricks naming rights at California Memorial Stadium and helmet branding in return for a deal that will make the university a shareholder in the AI company. The arrangement turns college sport into a…
Anthropic Watch
Anthropic is asking shareholders to approve a structure that would give its seven co-founders collective control of most corporate decisions, according to The Information. The proposal would hand the founders 50.1% of the voting power, despite their relatively small ownership stakes, just as the AI company prepares for…
Watch Desk
Documentation examples such as your-domain.com and yoursite.com are being used in malvertising campaigns, researchers at Manifold Security say. The practical warning is simple: a harmless-looking placeholder copied from code, documentation or an AI agent skill may now lead some visitors towards scams rather than a blank page.
Geoffrey Hinton Watch
Geoffrey Hinton has urged governments to put stronger safeguards around artificial intelligence, comparing the technology to a fast car without brakes or a steering wheel. His warning matters because the same discussion is now widening from laboratory safety to global inequality, public services and who gets to shape…
Meta AI Watch
Meta chief executive Mark Zuckerberg has rejected calls for AI companies to coordinate a slowdown, arguing that each lab should pause or reduce development when its own systems appear unsafe. The position puts Meta against recent proposals from Anthropic and OpenAI for shared action on frontier AI, and…
Google DeepMind Watch
Google DeepMind and Wiz have launched Scan for Good, a programme using AI with human security researchers to find and help fix exposed systems across public services, critical infrastructure and non-profits. Its early case studies include a hospital alert system, a rail operator, a national archive and a…
AWS AI Watch
AWS has published a GPU-ready SageMaker container for WhisperX that can transcribe audio, identify speakers and add word-level timestamps. The practical appeal is straightforward: teams can turn messy conversations into searchable, captionable and analysable records without building the whole speech pipeline from scratch.
Meta AI Watch
A security researcher says Meta’s Muse desktop AI client for macOS contained an unpatched flaw that let locally running software redirect voice-dictation traffic to an attacker-controlled server. Meta has since deployed a hotfix, according to InfoQ, but the episode shows how an AI agent’s broad permissions can turn…
Cohere Watch
Canada’s AI sovereignty debate is shifting away from the tidy idea of building everything at home. Cohere and Microsoft Canada argue that the more useful question is which parts of the international AI supply chain the country cannot afford to lose control over.
Watch Desk
Employees at Anthropic, OpenAI and Google DeepMind are pouring millions into US political campaigns and political action committees that support AI regulation, according to The San Francisco Standard. The movement is becoming an organised counterweight to more than $100m in industry spending aimed at shaping the debate, making…
Watch Desk
Brahma AI, the artificial-intelligence unit of VFX company DNEG, has raised $150m at a post-money valuation of $2bn, according to CNBC. The funding gives a relatively clear signal about where generative video is heading next: enterprise production, not just viral clips and prompt contests.
Watch Desk
Canonical will begin publishing Ubuntu kernel updates weekly from 28 September, bringing security fixes to users faster as the company cites a sharp rise in vulnerabilities uncovered through automated AI bug discovery. The practical trade-off is clear: quicker patches, with some users able to access less-tested fixes ahead…
Yann LeCun Watch
Yann LeCun has dismissed claims that AI could kill everyone within three years, while other technology leaders argue the immediate danger lies in cyberattacks and misuse. The disagreement matters because the AI safety debate is not one argument, but several wearing the same oversized coat.
Watch Desk
Axios reports that allies of Donald Trump are targeting Anthropic chief executive Dario Amodei as a symbol of AI “doomerism”, turning arguments about AI safety into a sharper political fight. The dispute matters beyond one executive’s reputation: it could affect Anthropic’s relationship with the White House and the…
Watch Desk
Professor B Ravindran of IIT Madras argues that fears about AI ending humanity often skip the crucial middle steps: access, permissions, testing decisions and human control. His interview offers a useful counterweight to the current slowdown debate, without pretending frontier AI is harmless.
Watch Desk
Researchers at MIT’s Senseable City Lab are arguing that visual AI could help planners understand traffic, pollution, public space and greenery across entire cities, while warning that the same systems can deepen surveillance and social bias. The useful point is not that computers can now look at more…
