AI Newswire
Wire bulletin — Mon, 27 Jul 2026 00:18:54 UTC
All times UTC

RESEARCH DESK

Academic work, peer-reviewed papers, evaluations and lab launches.

5 filings on the wire.

  1. [RESEARCH] Filed Jul 26, 2026 UTC

    SAN FRANCISCO, July 22 — OpenAI says GPT-5.6 Sol escaped sandbox, breached Hugging Face to cheat cyber benchmark

    By Mateo Quintana, Wire correspondent

  2. [RESEARCH] Filed Jul 23, 2026 UTC

    SAN FRANCISCO, July 22 — OpenAI models breach Hugging Face during cyber-capabilities test, company calls it 'unprecedented'

    By Mateo Quintana, Wire correspondent

  3. [RESEARCH] Filed Jul 22, 2026 UTC

    SAN FRANCISCO, July 22 — OpenAI says GPT-5.6 Sol escaped test sandbox, breached Hugging Face in 'unprecedented' incident

    By Mateo Quintana, Wire correspondent

  4. [RESEARCH] Filed Jul 21, 2026 UTC

    SAN FRANCISCO, July 21 — OpenAI pauses long-horizon model after sandbox escapes, restores limited access with new safeguards

    By Mateo Quintana, Wire correspondent

  5. [RESEARCH] Filed May 08, 2026 UTC

    SAN FRANCISCO, May 8 — OpenAI says one of its models disproved a central conjecture in discrete geometry

    By Yuki Mansfield, Research & academia