Safety News tracker.

1. SecuritySecurityOctober 15, 20264 min read

   Anthropic gives METR employee-like access to assess AI safety

   Anthropic says six METR employees now have deep internal access to evaluate its models and safety practices, with publication rights subject to narrow redactions.anthropic.com
2. SecuritySecurityOctober 15, 20264 min read

   Opsin raises $21 million to monitor AI agents in company systems

   Business Insider reports that AI security startup Opsin raised $21 million to help companies track agent activity, data access and risky behavior.BUbusinessinsider.com
3. SecuritySecurityOctober 15, 20264 min read

   Cohere proposes acquiring autonomous AI security firm AISLE

   Cohere says it has agreed to acquire AISLE, which develops AI systems for finding and helping fix previously unknown software vulnerabilities.Ccohere.com
4. SecuritySecurityOctober 15, 20264 min read

   Bank of America hires 1,000 contractors to address AI-related software vulnerabilities

   Bank of America CEO Brian Moynihan said the bank hired 1,000 external consultants to patch software vulnerabilities exposed by powerful AI models, according to Business Insider.BUbusinessinsider.com
5. SecuritySecurityOctober 15, 20264 min read

   Paper finds finite AI safety tests can fail against adaptive adversaries

   An arXiv paper argues that embodied and tool-using AI systems can behave safely through certification, then intervene after detecting that evaluation has ended. Its simulations and randomized AgentDojo trials report more post-certification security violations under this strategy, while permanent revocation and live…aXarxiv.org
6. SecuritySecurityOctober 14, 20264 min read

   Block says OpenAI-powered security tools cut investigation time

   Block says its security team uses OpenAI technology to accelerate vulnerability discovery, detection tuning and incident investigations while retaining human review.openai.com
7. SecuritySecurityOctober 14, 20264 min read

   DHS plans no-bid AI surveillance platform for ICE data

   Project Salt Box reports that DHS plans to award Pen-Link a possible five-year contract for an AI-assisted platform combining wiretap, phone, location, financial, email and social-media records for Homeland Security Investigations.PRprojectsaltbox.com
8. PolicyPolicyOctober 14, 20264 min read

   New York-led coalition seeks safeguards in Character.AI bankruptcy sale

   The River Reporter reports that New York Attorney General Letitia James and 39 other attorneys general urged bankruptcy-court protections for Character.AI users, especially children, before the chatbot company’s data and platform can be sold.RIriverreporter.com
9. SecuritySecurityOctober 14, 20264 min read

   UK AI startup Augur gets government backing and TfL public-safety project

   Resilience Media reports that Augur has received undisclosed strategic funding from the UK’s Sovereign AI Fund and begun a Transport for London research project exploring AI detection of sexual offending on London buses.REresiliencemedia.co

Every verified story on this topic in the order it happened, grouped by day (UTC). Updated canonical stories are flagged.

1. Friday, October 162026
   1. 10:30 AM UTCSecurity
      China displays AI-guided interceptor drones at Beijing forumVerified source: ua.news
2. Thursday, October 152026
   1. 11:46 PM UTCSecurity
      ChangeNOW launches NOW AI security lab after fixing website integration vulnerabilityVerified source: techbullion.com
   2. 8:04 PM UTCSecurity
      Harmonic Security raises $50 million for workplace AI governanceVerified source: scworld.com
   3. 5:24 PM UTCSecurity
      Anthropic gives METR employee-like access to assess AI safetyVerified source: anthropic.com
   4. 1:06 PM UTCSecurity
      Opsin raises $21 million to monitor AI agents in company systemsVerified source: businessinsider.com
   5. 12:00 PM UTCSecurity
      Cohere proposes acquiring autonomous AI security firm AISLEVerified source: cohere.com
   6. 5:24 AM UTCSecurity
      Bank of America hires 1,000 contractors to address AI-related software vulnerabilitiesVerified source: businessinsider.com
   7. 4:21 AM UTCSecurity
      Paper finds finite AI safety tests can fail against adaptive adversariesVerified source: arxiv.org
3. Wednesday, October 142026
   1. 11:18 PM UTCSecurity
      Block says OpenAI-powered security tools cut investigation timeVerified source: openai.com
   2. 8:07 PM UTCSecurity
      DHS plans no-bid AI surveillance platform for ICE dataVerified source: projectsaltbox.com
   3. 7:02 PM UTCPolicy
      New York-led coalition seeks safeguards in Character.AI bankruptcy saleVerified source: riverreporter.com
   4. 1:39 PM UTCSecurity
      UK AI startup Augur gets government backing and TfL public-safety projectVerified source: resiliencemedia.co
   5. 10:31 AM UTCSecurity
      Nvidia pledges $250m in computing power for UK AI safety researchVerified source: theguardian.com
   6. 5:33 AM UTCSecurity
      Chainguard launches AI-powered harness for verified vulnerability huntingVerified source: unite.ai
   7. 2:15 AM UTCSecurity
      Google DeepMind and Singapore announce on-device scam detection modelVerified source: deepmind.google
   8. 1:44 AM UTCSecurity
      Raonsecure gets Korean AI certification for OneShield security platformVerified source: biz.chosun.com
   9. 12:39 AM UTCSecurity
      Georgia Tech researchers test probes for early signs of harmful AI behaviorVerified source: news.gatech.edu
4. Tuesday, October 132026
   1. 11:57 PM UTCSecurity
      OpenAI engineer charged after covert recorder found at officeVerified source: justice.gov
   2. 6:02 PM UTCSecurity
      Microsoft outlines four controls for AI agents changing production networksVerified source: commandline.microsoft.com
   3. 5:30 PM UTCPolicy
      New York City Council hears labor concerns as it proposes AI safeguardsVerified source: wnylabortoday.com
   4. 3:50 PM UTCSecurity
      UK and Canada renew partnership on frontier AI securityVerified source: gov.uk
   5. 2:45 PM UTCSecurity
      Hadrian raises $40m to expand agentic security platformVerified source: aimagazine.com
   6. 3:59 AM UTCSecurity
      Anthropic finds early signs of AI self-replication capabilityVerified source: anthropic.com
   7. 2:52 AM UTCSecurity
      AI agents move closer to crypto trading—and expose new security risksVerified source: fortune.com
5. Monday, October 122026
   1. 11:36 PM UTCSecurity
      Five Eyes nears agreement on shared military AI backboneVerified source: rnz.co.nz
   2. 11:23 AM UTCSecurity
      Apate deploys AI bots to waste scammers' time and gather intelligenceVerified source: wired.com
6. Sunday, October 112026
   1. 11:34 PM UTCPolicy
      Microsoft president urges UK to lead global AI safety rulesVerified source: bbc.com
   2. 7:21 PM UTCSecurity
      Bill Gates warns AI‑driven cyberattacks will dominate security challenges this yearVerified source: sudouest.fr
   3. 6:50 PM UTCSecurity
      Japan issues nationwide cyber warning as AI‑assisted attacks surgeVerified source: the420.in
   4. 5:48 PM UTCSecurityUpdated
      Bill Gates warns AI‑driven cyberattacks could surge within a yearVerified source: portafolio.co
   5. 1:34 PM UTCSecurity
      Booz Allen tests show AI agents can breach industrial control systemsVerified source: news.lavx.hu
   6. 11:28 AM UTCIndustry
      AI safety evaluators gain prominence as labs face internal and external pressureVerified source: cnbc.com
   7. 9:20 AM UTCEnterprise
      University of Maine System signs two‑year ChatGPT Edu contract with OpenAIVerified source: mainepublic.org
   8. 6:34 AM UTCSecurity
      Researchers use explainable AI to surgically break LLM safety filtersVerified source: bioengineer.org
   9. 2:51 AM UTCProduct
      ONLYOFFICE 10.0 adds context‑aware AI tools and modular apps to its open‑source suiteVerified source: bastillepost.com
   10. 1:48 AM UTCSecurity
       Hackers use Google ad and Bing redirect chain to deliver fake Claude installerVerified source: pushsecurity.com
7. Saturday, October 102026
   1. 11:36 PM UTCSecurity
      Critical vulnerability in GitLab AI Gateway affects Duo Agent Platform usersVerified source: secnews.gr
   2. 9:28 PM UTCPolicy
      Microsoft CEO says companies should treat AI models as trust risksVerified source: businessinsider.com
   3. 4:45 PM UTCPolicy
      Senator Warner introduces AI Risk Management and Security ActVerified source: quiverquant.com
   4. 9:37 AM UTCSecurity
      Met Police dismiss volunteer chief after Palantir AI probe finds misconductVerified source: bbc.com
   5. 3:52 AM UTCSecurity
      Kenton County Police investigate AI-driven sextortion scam targeting teenVerified source: wlwt.com
   6. 2:18 AM UTCSecurity
      Super Micro contractor pleads guilty in AI server diversion schemeVerified source: straitstimes.com
   7. 1:47 AM UTCSecurity
      Anthropic reports new AI misbehavior on government sitesVerified source: news.bloomberglaw.com
   8. 1:16 AM UTCSecurity
      Anthropic blocks live internet access for all internal AI evaluations after unintended model actionsVerified source: anthropic.com
   9. 12:13 AM UTCSecurity
      OpenAI pauses GPT-6.1 Astra release over security concernsVerified source: ibtimes.com
8. Friday, October 92026
   1. 11:41 PM UTCSecurity
      Anthropic says its AI agents tried to access multiple government websitesVerified source: nytimes.com
   2. 11:09 PM UTCSecurity
      Ukraine drones knock out Russian AI data center linked to YandexVerified source: arstechnica.com
   3. 9:29 PM UTCSecurity
      Anthropic AI model submits false homicide tip to Philadelphia policeVerified source: theverge.com
   4. 8:55 PM UTCIndustry
      Anthropic targets record $100 billion IPO in NovemberVerified source: wikileaks-kr.org
   5. 6:18 PM UTCSecurity
      Ethereum and Bitcoin projects seek early access to Anthropic's AI scannerVerified source: finance.biggo.com
   6. 5:15 PM UTCSecurity
      OpenAI fires three safety researchers over alleged breach of trustVerified source: business-standard.com
   7. 3:41 PM UTCPolicy
      Democrats unveil AI safety and liability frameworks ahead of midterm electionsVerified source: politico.com
   8. 1:35 PM UTCSecurity
      Anthropic launches free OSS Scanner using Claude Mythos to auto‑scan open‑source projectsVerified source: timesofindia.indiatimes.com
   9. 1:04 PM UTCSecurity
      SemiAnalysis report finds 3.6% of Chinese AI releases have public safety testsVerified source: kfgo.com
   10. 4:32 AM UTCPolicy
       South Africa urges global treaty to regulate military AI and autonomous weaponsVerified source: citizen.co.za
   11. 3:14 AM UTCSecurity
       Chinese developer closes ARTEX AI agent after South Korean bank hacksVerified source: straitstimes.com
9. Thursday, October 82026
   1. 11:03 PM UTCPolicy
      Democratic lawmakers unveil draft AI safety bill with shutdown authorityVerified source: politico.com
   2. 10:00 PM UTCSecurity
      Tenable launches Exchange Inspector to vet AI agents with OpenAI GPT‑Cyber modelsVerified source: securityboulevard.com
   3. 5:16 PM UTCIndustry
      Arena raises $200M and launches AI Alignment IndexVerified source: cryptobriefing.com
   4. 1:04 PM UTCSecurity
      Zenity researchers expose privileged AWS Bedrock agent role that lets a compromised agent hijack a fleetVerified source: labs.zenity.io