aibrief.fyi
AI news, with memory.
Tuesday, August 25, 2026
Topic archive

Safety

  1. Researcher says Microsoft AI images in Paint and Photos include watermarks tied to user IDs
  2. Alabama opens investigation into OpenAI model hacking incident involving Hugging Face
  3. OpenAI urges California to strengthen SB 53 after previously opposing the bill
  4. Study finds frontier AI labs disclose little about rogue-model containment plans
  5. Report highlights allegedly racist safety advice from a Google AI system
  6. OpenAI launches Private Safety Processing with zero data retention for enterprise customers
  7. OpenAI reportedly halted training on an advanced model after detecting concerning behavior
  8. European Central Bank warns AI-driven market exuberance could amplify a broader financial selloff
  9. Researchers say OpenAI revoked access for some participants in its Trusted Access for Cyber program
  10. Coders quickly report ways to evade Anthropic’s Claude text watermarks
  11. Wired reconstructs Flock Safety's newer police AI investigation tool already in use
  12. OpenAI hardens chain-of-thought security monitoring, increasing overhead for some workloads
  13. Robin Williams' family revives his Instagram account to respond to AI likeness misuse
  14. Z.ai releases open-weight cybersecurity models with dual-use bug-finding capabilities
  15. Report alleges Amazon destroyed rare books to obtain AI training data
  16. OpenAI reportedly disbanded its preparedness team
  17. TechCrunch report details alleged misuse of Grok to create explicit imagery from a childhood photo
  18. OpenAI reportedly alerted the FBI over disturbing ChatGPT conversations linked to a Goldman Sachs analyst
  19. Writer launches Palmyra X6 and updates its enterprise agent orchestration and governance stack
  20. Anthropic study shows Claude-based agents can escalate into sabotage on shared servers
  21. Flock Safety adds safeguards to AI surveillance tools after backlash
  22. ShieldFont launches a font designed to poison AI web scraping while remaining readable to humans
  23. Report says AI-driven agents targeted Taiwan’s nuclear safety agency in a cyberattack
  24. Supply-chain attack on AI package reportedly led to terabytes of stolen credentials
  25. Report says OpenAI ethics lead Miles Brundage has left the company
  26. Bernie Sanders warns AI company leaders against building systems humans cannot control
  27. Report alleges Google hiring AI discarded some qualified job applications
  28. Researchers disclose a now-fixed Zoom screen-sharing flaw that could let call participants take over another device
  29. Farmer says AI-generated farming advice led to loss of 25 acres of sesame crop
  30. Protesters are arrested after entering OpenAI's Washington lobbying office
  31. OpenAI expands Daybreak with a higher-access tier for its cybersecurity model
  32. DEF CON water-utility security project adds providers and expands digital-twin and AI capabilities
  33. Kimsuky is reportedly using local LLMs to enhance phishing operations
  34. Reports describe an OpenClaw AI agent manipulating a gym waitlist system
  35. Anthropic to enable Claude Code's auto mode by default
  36. Researchers use AI to design 16 new viruses in experiment with biosecurity implications
  37. Researchers say Moonshot AI’s Kimi escaped a misconfigured cybersecurity testing sandbox
  38. New Orleans deploys AI to assist 911 emergency call handling
  39. Meta says its AI showed hacking-related behavior
  40. Study finds human reviewers miss about one-third of risky AI coding-agent requests
  41. Check Point researchers report security flaws in AI agent frameworks ahead of Black Hat presentation
  42. CISA says critical Langflow remote-code-execution flaw is under active exploitation
  43. PwC reportedly published an AI report containing hallucinated citations and errors
  44. ChatGPT appears to refuse prompts asking it to imitate living authors' styles
  45. OpenAI says test agent used exposed logins to access at least four public services
  46. Glow emerges from stealth focused on endpoint security risks from enterprise AI agents
  47. Suno user data breach reportedly exposed information on 55 million accounts
  48. OpenAI safety leader Johannes Heidecke is leaving the company
  49. OpenAI launches GPT-5.6 and a broader new model family
  50. Anthropic launches Claude Sonnet 5 as a lower-cost model for agentic workloads
  51. Wired reports Meta contractors tested rival chatbots by posing as teenagers in high-risk conversations
  52. OpenAI unveils GPT-5.5-Cyber update and launches Patch the Planet bug-fixing initiative
  53. OpenAI launches Lockdown Mode in ChatGPT to limit prompt-injection data exposure
  54. Hackers reportedly exploited Meta's AI support chatbot to take over Instagram accounts
  55. Illinois legislature passes AI safety bill requiring third-party compliance checks
  56. OpenAI says a code security incident led to limited employee-device data theft
  57. Exaforce raises $125 million Series B for AI-driven real-time cybersecurity
  58. OpenAI adds a Trusted Contact safeguard in ChatGPT for possible self-harm situations
  59. Braintrust confirms cloud breach and urges all customers to rotate API keys