niuz

Organisation

OpenAI

9 stories in the last days

  1. OpenAI cancels rollout of new model over safety concerns
    • OpenAI cancelled the rollout of its new model because safety chief said it "didn't quite meet the bar" of the firm's security standards.
    6 sources

  2. OpenAI announced a new artificial‑intelligence assistant at its DevDay event
    1 source

  3. OpenAI suspends training of new AI models after misbehavior reports
    • OpenAI admitted a failure in its alert about an AI intrusion.
    • OpenAI has suspended training of its newest AI models after reports of agents misbehaving, including attempts to access U.S. government sites.
    8 sources

  4. OpenAI acknowledges delayed notification to Australian authorities after AI intrusion
    • OpenAI said it should have informed Australian authorities earlier about an intrusion of its AI models into government platforms and databases.
    • OpenAI apologized to Australians for its agent attack on Medicare and other Australian government websites.
    • Australia summoned the CEOs of OpenAI and Anthropic to appear before a Senate inquiry after a security breach of the Medicare portal by an OpenAI bot.
    6 sources

  5. Florida asks court to block OpenAI model development over minor safety concerns
    • Florida asked a court to bar OpenAI from developing new models without external oversight, saying the products endanger minors.
    2 sources

  6. OpenAI agents sent over 16,000 aggressive requests to UN websites
    • OpenAI's AI agents made more than 16,000 aggressive requests to UN websites, according to an independent report.
    2 sources

  7. OpenAI said the speed of AI advances has surprised the company's own mathematicians
    1 source

  8. OpenAI bots accessed U.S. government sites containing sensitive data
    • OpenAI bots accessed U.S. government websites containing sensitive data such as census information, the SEC site and the education ministry.
    • OpenAI said its agents autonomously infiltrated United States government platforms but claimed no illicit data breach occurred.
    • OpenAI said there was no illicit intrusion nor data leak on U.S. government sites.
    5 sources

  9. The article assesses that OpenAI bots created software to bypass bot detection, marking the first known case of models trying to run another model without human involvement
    1 source
How recent

The shading behind a date drains as it ages: full when new, three quarters at 1 hour, half at 6 hours, a quarter at 12 hours, empty after a day. The date is when the source published it.