DeepMind AI agents reportedly weaken legacy 1024-bit RSA

A quiet release day turns on security: OpenAI widens a red-team firm's model access, and Copilot CLI's autopilot stops halting on permission prompts.

Nowline NOV 8 4:00 PM banner

Top AI stories from the last hour

Top AI stories from the last hour

Copy markdown

  • AI cryptanalysis reaches legacy 1024-bit RSA

    Google DeepMind's AI agents reportedly surfaced two weaknesses in legacy 1024-bit RSA — including a signature-forgery path and improved factoring — according to two outlets; 2048-bit and larger keys are unaffected. If any 1024-bit keys or certificates still linger in your stack, rotate them: the theoretical safety margin just narrowed.

  • OpenAI opens Daybreak Red to a red-team firm

    OpenAI is widening security startup Hacktron's access to its frontier models via the Daybreak Red program, and the two will jointly probe how OpenAI's services interact with third-party platforms — the same integration surfaces your agents call. OpenAI says it found Hacktron through its Security Bug Bounty. Read it as more adversarial testing (and fixes) landing on tool and connector behavior.

  • Copilot CLI autopilot stops halting on permissions

    GitHub Copilot CLI 1.0.108-0 changes Autopilot with Assisted Permissions so it no longer stops at a permission prompt: it tries another approach and reports the actions it couldn't take. Unattended runs stall less — just read the "couldn't do" summary before you trust the output.