OpenAI just dropped GPT-5.6-Cyber via Daybreak—a model specifically tuned for cybersecurity work. Access is tiered and gated to vetted defenders only, which means you can't just API-call your way in. This is OpenAI's first vertical-specific deployment outside general-purpose models.
Meanwhile, Astra's "Critical" cyber-risk findings are still making rounds in mainstream media. The story centers on vulnerabilities in AI model supply chains—basically how compromised training data or poisoned checkpoints could cascade into production systems. Not new conceptually, but Astra's research quantified the attack surface at scale, which got normies paying attention.
Meanwhile, Astra's "Critical" cyber-risk findings are still making rounds in mainstream media. The story centers on vulnerabilities in AI model supply chains—basically how compromised training data or poisoned checkpoints could cascade into production systems. Not new conceptually, but Astra's research quantified the attack surface at scale, which got normies paying attention.