tech
Anthropic says Mythos can turn software patches into exploits in minutes
Advanced models are just as effective at rapidly weaponizing flaws that defenders already know about.
TL;DR
- Anthropic's Mythos AI can turn newly disclosed software vulnerabilities into working exploits in hours.
- Mythos generated a proof-of-concept exploit for a Windows kernel vulnerability in 31 minutes.
- It successfully caused a 'blue screen of death' in 18 out of 21 tested kernel bugs.
- Mythos created 8 distinct exploits for Windows kernel bugs, with the longest taking 5.7 hours.
- The AI also developed 8 working code-execution exploits for 18 Firefox security patches.
- Advanced AI models may be as effective at weaponizing known flaws as they are at finding new bugs.
- This capability could significantly shorten the 'patch gap' between vulnerability disclosure and widespread patching.
- Open-source models and competitors like GPT-5.5-Cyber are also demonstrating similar bug-finding and weaponization capabilities.
- The cost for Mythos to generate Windows privilege-escalation exploits was estimated at around $15,700 in API credits.
- The Trump administration is implementing an AI security executive order to assess national security risks from advanced AI models.