โ ๏ธ OpenAI slows Astra after it hits first-ever 'critical' cybersecurity risk level
OpenAI flagged its upcoming Astra model as potentially reaching the "critical" cybersecurity tier under its Preparedness Framework โ a first for any of its models. Internal tests showed Astra can craft zero-day exploits and devise end-to-end cyberattack strategies on hardened targets without human help. OpenAI has slowed development and plans to loop in government agencies and independent AI safety groups before release. How worried should we actually be about AI that can independently hack critical infrastructure?
โก 42 buzzing
573 views
๐ 0 subscribed