Anthropic's Claude Mythos outpaces GPT-5.5 and solves 'Cooling tower' challenge
Technology
Anthropic's new artificial intelligence (AI), Claude Mythos, has pulled ahead of OpenAI's GPT-5.5 in recent cybersecurity challenges, according to the UK AI Security Institute (AISI).
Mythos even cracked the tricky "Cooling Tower" problem (a first for any model) about a month after its initial release, via Project Glasswing, the cybersecurity testing alliance, a collaborative effort to boost digital security.
AISI: Cyber task lengths doubling
AISI found that these AI models are improving at lightning speed, with the length of cyber tasks AI models could complete having doubled every 4.7 months since late 2024.
Even with strict testing limits, both Mythos and GPT-5.5 scored high marks, hinting they are capable of even more.