AISLE
4 mentions across all digests
AI security company operating a vulnerability discovery and remediation system that discovered 15 CVEs in OpenSSL, 5 in curl, and 180+ total across 30+ projects since mid-2025.
On Anthropic’s Mythos Preview and Project Glasswing
Anthropic withheld Claude Mythos Preview from public release, instead deploying it through Project Glasswing to proactively discover and patch software vulnerabilities before competitors can exploit them.
Anthropic's Claude Mythos Launch Is Built on Misinformation
Anthropic's Claude Mythos launch narrative around vulnerability discovery overstates model differentiation; investigation finds 3.6B-parameter models replicate key results, with Mythos's real advantage lying in engineering scaffold rather than raw capability.
Mythos and Cybersecurity
Anthropic's Claude Mythos Preview weaponizes 181 Firefox exploits under restricted access to 50 vendors, but Schneier warns this may leave industrial and medical infrastructure dangerously undefended while skilled adversaries exploit the model elsewhere.
Small models also found the vulnerabilities that Mythos found
AISLE researchers show small open-weight models replicate Anthropic's Mythos vulnerability-finding capabilities at 1/100th the cost, proving AI security breakthroughs depend on methodology and expertise rather than frontier model scale.