analiza aiqbench
Microsoft is establishing a strict behavioral framework for its advanced AI models that explicitly forbids autonomous cyberattacks or self-escalation of privileges, prioritizing human control over system capabilities. This policy mandates that agents remain interruptible and transparent, with absolute constraints preventing the generation of exploit code or attack tools regardless of user prompts. The distinction drawn between defensive cybersecurity assistance and offensive capabilities sets a new precedent for how enterprise AI safety protocols will be enforced at the model level.
Microsoft Bans Its AI Models From Launching Cyberattacks or Escalating Their Own Access
Microsoft has published a draft Humanist AI Code of Conduct that would prohibit its in-house MAI models from launching cyberattacks, supplying operational attack capabilities, or increasing their own privileges. The rules also require agents to remain interruptible, transparent, and confined to human-assigned permissions and objectives. The document, released for a six-week public consultation, is intended […] The post Microsoft Bans Its AI Models From Launching Cyberattacks or Escalating Their Own Access appeared first on Cyber Security News.
Microsoft has published a draft Humanist AI Code of Conduct that would prohibit its in-house MAI models from launching cyberattacks, supplying operational attack capabilities, or increasing their own privileges. The rules also require agents to remain interruptible, transparent, and confined to human-assigned permissions and objectives. The document, released for a six-week public consultation, is intended to become the primary behavioral framework for models developed by Microsoft AI.…
Extras afișat conform termenilor RSS ai publicației · 67 cuvinte · textul integral și figurile sunt pe site-ul publicației
Subiecte conexe
Problems of Proliferation
LessWrong · 29 sept. 2026
[AINews] Meta Muse Wonder & Wonder Full: MSL’s Claude 5.5-tier Frontier Model
Latent Space · 13 oct. 2026
[AINews] Meta Connect 2026: Muse glasses, voice, video, and Charm
Latent Space · 24 sept. 2026
Digital States 2026: AI Moves From Experiment to Infrastructure
govtech.com · 24 sept. 2026