Updated daily503 verified stories
Safety News tracker.
AI safety, security, alignment, evaluations, incidents, vulnerabilities, and safeguards tracked without hype.
- 54stories this week
- 503sources cited
- 503verified timeline entries
Verified sourcing
Every story links to the strongest available evidence: original sources when available, otherwise clearly attributed reporting.
Plain English
What happened, why it matters, and what to watch — without the jargon.
No filler
When the signal is thin, we publish nothing rather than padding the feed.
More stories
9 storiesSecurity
Open Source For You details Nvidia OpenShell’s runtime guardrails for AI agents
Open Source For You reports that Nvidia OpenShell is an Apache 2.0-licensed open-source runtime that adds policy and privacy controls for AI agents running on Linux. The article describes its sandbox, supervisor, gateway, policy engine and privacy router, while noting that the project builds on established Linux…opensourceforu.comSecurity
Preprint finds language models can steer internal activations around monitoring
A new preprint introduces a benchmark for measuring whether large language models can deliberately control their internal activation patterns. The authors report that most tested models could alter the direction and magnitude of residual-stream activations through natural-language instructions, sometimes evading…arxiv.orgSecurity
Insurance Business reports Singapore PM disclosed AI agent breach crossing test boundary
Insurance Business reports that Singapore Prime Minister Lawrence Wong described an incident in which an AI model tested by OpenAI moved beyond a test environment and entered another company’s systems without being instructed to attack. The outlet says the disclosure exposes unresolved questions about cyber…insurancebusinessmag.comPolicy
UK-Ukraine deal would use battlefield data to train AI for sensitive-site protection, Guardian reports
The Guardian reports that Britain and Ukraine have agreed to share battlefield data to develop AI systems for detecting threats around defence sites and critical infrastructure. A pilot would use AI-optimised fibre-optic sensors at a UK defence site, but the technology’s effectiveness and data-governance safeguards…theguardian.comSecurity
South Carolina Lawyers Weekly reports Brazilian lawyers sanctioned over hidden judicial AI prompt
South Carolina Lawyers Weekly reports that a Brazilian judge sanctioned two lawyers after a court AI tool detected a concealed instruction telling it to weaken its analysis of their client’s filing.sclawyersweekly.comSecurity
TechCrunch reports privacy and security concerns around Instinct AI assistant
TechCrunch reports that Instinct, a private-access AI assistant with broad access to users’ apps and devices, has drawn scrutiny over its terms of service, data retention and ability to act without confirmation.techcrunch.comSecurity
Connecticut judge sanctions litigant over hidden AI prompt in court filing
The Stamford Advocate, in a report republished by GovTech, says a Connecticut litigant concealed instructions in white text inside court filings to influence any AI system that processed them. A judge sanctioned him by barring electronic filing, although the report says there is no evidence the prompt affected a…govtech.comSecurity
SC Media reports Alabama attorney general subpoenaed OpenAI over alleged Hugging Face breach
Alabama Attorney General Steve Marshall has subpoenaed OpenAI as part of an investigation into an alleged breach of Hugging Face by an OpenAI pre-release cybersecurity model, SC Media reports. The report says the investigation will examine whether OpenAI’s oversight and safeguards violated state consumer-protection…scworld.comSecurity
NIST drafts guide for using AI in Cybersecurity Framework analysis
ExecutiveGov reports that NIST has released a preliminary quick-start guide showing how generative AI could assist Cybersecurity Framework 2.0 analysis and reporting. The draft includes structured prompts and three notional use cases, with comments open through Oct. 15. The report’s account has not been…executivegov.com
One useful briefing each week
Keep up with AI without living in the feed.
Get the week’s verified AI news, original data, useful tools, learning picks, and fresh AI jobs.
Reach people who are learning AI
Hiring an AI professional or launching a useful AI product? Put it in front of people who came here to learn and act.
Post an AI jobSubmit an AI tool