OpenAI and Anthropic Hide Cyber-Hacking AI Because Nobody Wants Apocalypse VIP Pass
KEY POINTS
- •OpenAI introduced the GPT-5.3-Codex, its most cyber-capable model, in February with a $10 million API credit program for select users.
- •Anthropic announced a restricted rollout of its hacking-capable Mythos Preview model to chosen cybersecurity companies to prevent misuse.
- •Experts warn AI with hacking abilities is already out and unstoppable, likening limited releases to decades-old software vulnerability disclosures.
OpenAI, following Anthropic’s lead, is treating its new cyber-savvy AI like a nightclub with a very exclusive guest list: only select companies get the golden ticket. After unleashing GPT-5.3-Codex in February and pledging a $10 million bribe—er, API credit—to 'trusted' organizations, OpenAI is quietly prepping a ‘trusted access’ pilot, because apparently giving everyone a digital supervillain toolkit is so 2023. Remember Anthropic's Mythos locked tighter than Fort Knox? They vowed no public release for Mythos Preview but might soften up for future versions if guardrails show up. Meanwhile, top security experts are having a nervous breakdown, reminding us this genie’s already out: AI now finds holes in code like your ex finds new reasons to ghost you. Prepare for the inevitable AI-equipped hackers who may disrupt electric grids or water utilities—because if there was ever a perfect sequel to The Matrix, this is it.
Share the Story
(1 of 3)Source: Axios | Published: 4/9/2026 | Author: Sam Sabin