Anthropic unveiled Project Glasswing Tuesday, offering its new AI model Mythos Preview to a select group of tech firms for vulnerability hunting. The system has already uncovered thousands of zero‑day bugs across major OSes and browsers, according to the company. Anthropic says the model now rivals all but the most elite human security researchers. Its developers claim the AI can rewrite exploit code in minutes.

Apple joins the initial roster alongside AWS, Google, Microsoft, NVIDIA and others, gaining access to Mythos Preview for defensive work on Safari and its suite of operating systems—iOS, iPadOS, macOS, watchOS, tvOS and visionOS. Anthropic is allocating up to $100 million in usage credits, after which partners will pay per query. The credits cover early‑stage testing, after which pricing will reflect compute intensity.

By funneling advanced AI into its bug‑bounty pipeline, Apple hopes to patch flaws faster than attackers can weaponize them, tightening security across billions of devices. The closed‑beta approach keeps Mythos Preview out of malicious hands while giving partners a powerful tool against an accelerating threat surface. Anthropic’s move signals a shift toward AI‑driven defensive collaboration. It also reduces reliance on external pen‑test services.