Published • loading... • Updated
OpenAI Says Upcoming Model Is so Capable It Requires Stronger Guardrails
OpenAI said Astra can find and exploit unknown security flaws, prompting tighter safeguards and limited access to its cybersecurity capabilities.
- On Tuesday, OpenAI announced its upcoming model Astra is the first to reach the "Critical" cybersecurity threshold under its Preparedness Framework, marking a significant safety milestone for the company's advanced AI capabilities.
- Astra can identify and exploit unknown software vulnerabilities without human guidance, allowing it to "chain" multiple exploits to penetrate well-protected systems, OpenAI VP of research Amelia Glaese told reporters.
- Following a multi-week pause in training to bolster defenses, OpenAI implemented stronger safeguards including training the model to refuse harmful cyber requests and respect safety restrictions, officials said.
- While OpenAI plans to release Astra "soon," access to advanced capabilities will be restricted using a new "misalignment monitor" that may inadvertently flag legitimate activity as unauthorized behavior.
- Partners in the Daybreak program—including Cisco, Cloudflare, and Palo Alto Networks—will receive early access to a less restricted version to help harden defenses before broader release.
Insights by Ground AI
14 Articles
14 Articles
OpenAI to limit access to Astra's most powerful cyber capabilities
·Arlington, United States
Read Full Article
OpenAI says upcoming model is so capable it requires stronger guardrails
·London, United Kingdom
Read Full Article
Coverage Details
Total News Sources14
Leaning Left3Leaning Right0Center5Last UpdatedBias Distribution62% Center
Bias Distribution
- 62% of the sources are Center
62% Center
L 38%
C 62%
Factuality
To view factuality data please Upgrade to Premium
