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
Utah Governor Spencer Cox has signed an order setting out how state agencies should bring AI into government. It assigns responsibilities for staff training, cybersecurity safeguards and human review of automated decisions affecting individual rights, KSL reports.
Read the report & discussion ↗Watch Desk
Mark Zuckerberg says Meta’s Llama 4 setback was partly an organisational mistake: he built its team on a model suited to large-scale product work, rather than the smaller, closely connected group he believes language-model research needs. His account points to a consequential shift in how Meta is organising…
Watch Desk
Researchers say they found 8,547 internet-facing systems linked to wind and solar sites across 35 European countries, including interfaces that can display operational data and controls. They told Reuters that they believed full control may have been possible at around 181 sites, putting a concrete security question to…
Microsoft AI Watch
Microsoft says AI-powered vulnerability scanning is likely to bring security teams more findings than their existing review processes were built to handle. Its advice to security leaders includes considering fixes for critical systems within 24 hours, rather than waiting for a convenient maintenance window.
Watch Desk
AI policy is beginning to shape the conversation around possible Democratic presidential contenders, with candidates split over how far to regulate frontier systems and data centres. A new Silver Bulletin analysis compares eight prospective contenders, showing that the debate is already broader than a simple choice between innovation…
Anthropic Watch
Anthropic has expanded its Cyber Verification Program, giving qualifying security teams access to cyber capabilities and reduced blocking safeguards through three tiers. The key change is a broader route to powerful tools for defensive work, with tighter checks as permitted activity moves towards authorised red-teaming and sensitive systems.
Watch Desk
A company can buy AI tools from different vendors and still depend on the same underlying models, APIs or computing providers. Swati Deepak Kumar, a Citi senior vice-president of IT, argues that this can leave important business decisions exposed to the same blind spot or behaviour change, even…
AWS AI Watch
AWS has published guidance on using ISO/IEC 42005:2025 to assess the impacts of AI systems. The useful point is that assessments can be fitted into existing risk reviews, with a lighter triage to identify where a fuller examination is needed.
Watch Desk
Google says research prototypes combining geospatial data and AI helped public health teams identify 48 settlements and more than 45,500 people at risk during the ongoing Ebola outbreak in the Democratic Republic of the Congo. The work points to a practical use for AI: helping responders find gaps…
Watch Desk
IBM and Red Hat say their Lightwell programme has fixed more than 400 previously unknown vulnerabilities in widely used Java libraries and opened a service for enterprise customers to request priority reviews and backported fixes. The announcement also ties the effort to risks from autonomous AI agents, though…
Watch Desk
Large language models can make historical OCR easier to read, but fluent corrections may alter spellings, names or passages that the page does not support. A new survey maps the main approaches and argues that preserving the source needs to be measured alongside transcription accuracy.
Mistral AI Watch
Mistral has opened a public preview of Mistral Large 4, nicknamed Le Chonk, a trillion-parameter multimodal model with a particular pitch for cybersecurity. The company plans to release its weights on 27 October, after more testing, giving customers a chance to try the preview before they can run…
Watch Desk
AI agents could one day guide people through legal problems, draft claims and negotiate with institutions, the Law Society warns. Its foresight study argues that accountability and meaningful human oversight need settling before such systems become a doorway into justice.
Meta AI Watch
Meta’s Oversight Board overturned the company’s decision to leave a deepfake of a Scottish councillor on Facebook, finding that it should have been removed under Meta’s Hateful Conduct rules. The Board also called for stronger friction, penalties and transparency around high-risk AI content.
Watch Desk
SignSplit has launched with a reported $400 million strategic seed commitment, pitching infrastructure for people to license personal data, creative work and likenesses to AI developers. The bigger question is whether signed consent and payment can become practical plumbing for AI training, rather than another promise attached to…
Meta AI Watch
Meta’s Ray-Ban smart glasses are gaining walking directions, alongside audio controls that let users balance sound between speakers or switch to mono. The changes make the glasses more useful for getting around and offer practical listening options, not just another setting to admire in a menu.
Watch Desk
New York City Council heard former AI-company employees warn that frontier AI could become difficult to control, while company representatives faced questions about catastrophic risks and possible local safeguards. The hearing also put concrete proposals, including whistleblower protections and third-party checks, into the debate.
Watch Desk
Elon Musk and Anduril co-founder Palmer Luckey are helping lead a Pentagon study of future warfare, including AI-powered weapons. Their defence businesses could benefit from some of the study’s recommendations, raising questions about how its work will be insulated from commercial interests.
Watch Desk
A new book by technology journalist Kevin Roose examines how rivalry, safety concerns and genuine excitement coexist inside leading AI companies. The New York Post’s account offers a vivid, if sometimes startling, look at the people shaping a race whose consequences reach well beyond the labs.
Watch Desk
Mi Familia en Acción has launched Nubi, a bilingual AI app offering locally tailored answers to voting questions across all 50 US states. The civic group says the tool draws on information it considers reliable rather than searching the open internet, a design choice that matters when an…
Watch Desk
The UK government will act on all 44 proposals for regulating healthcare AI, according to Resultsense. The shift puts more emphasis on monitoring AI medical devices after they reach patients, with draft guidance on product updates due by December.
