Frontier AI Models Tackle Hard Exploitation Tasks
Investors and academics react to Irregular post on AI model progress in security tasks.
Shaun Maguire compared the post to Crowdstrike's 2014 Putter Panda report and said leading labs take security seriously but few outsiders know what is coming. Pat Grady called it an excellent state of the union on AI security. Dean Meyer stated that no frontier model solved Irregular's hardest exploitation task in February, the best one succeeded occasionally for around $2,000 by April, and several succeeded reliably for around $20 by June. Mor Geva called the post an important read. Geoffrey Irving agreed that the path to defensive dominance will be rough.
𝗧𝗵𝗲 𝗘𝗻𝗱-𝗦𝘁𝗮𝘁𝗲 𝗙𝗮𝗹𝗹𝗮𝗰𝘆: 𝗪𝗵𝗲𝗿𝗲 𝗜𝘀 𝗔𝗜 𝗦𝗲𝗰𝘂𝗿𝗶𝘁𝘆 𝗚𝗼𝗶𝗻𝗴? Frontier AI models had a giant performance gain in coding in the Fall of 2025 Then with cybersecurity in April This is now happening with open-weight models We are optimistic about the…
Reactions
11 postsDM 
Dean Meyer@DeanMeyerrrFebruary: no frontier model could solve Irregular's hardest exploitation task. By April: the best one could, occasionally, for ~$2,000. By June: several could, reliably, for ~$20. Over the next year, we will likely enter a period where AI's offensive capability outpaces…
- likes: 70
- replies: 9
- bookmarks: 25
- reposts: 11
MG Mor Geva@megamor2Important read from @dan_lahav @Irregular about the state of AI Security and where it is headed
- likes: 16
- replies: 0
- bookmarks: 7
- reposts: 0
SM Shaun Maguire@shaunmmaguireDan and the @Irregular team live at the frontier of AI security Reading this gave me the same feeling I had w/ Crowdstrike’s “putter panda” report in 2014 The leading labs are taking security seriously—but too few outside of them know what’s coming Security is about to change
- likes: 132
- replies: 10
- bookmarks: 60
- reposts: 16
DL Dan Lahav@dan_lahav𝗧𝗵𝗲 𝗘𝗻𝗱-𝗦𝘁𝗮𝘁𝗲 𝗙𝗮𝗹𝗹𝗮𝗰𝘆: 𝗪𝗵𝗲𝗿𝗲 𝗜𝘀 𝗔𝗜 𝗦𝗲𝗰𝘂𝗿𝗶𝘁𝘆 𝗚𝗼𝗶𝗻𝗴? Frontier AI models had a giant performance gain in coding in the Fall of 2025 Then with cybersecurity in April This is now happening with open-weight models We are optimistic about the…
- likes: 313
- replies: 52
- bookmarks: 316
- reposts: 58
DM Dean Meyer@DeanMeyerrrFebruary: no frontier model could solve Irregular's hardest exploitation task. By April: the best one could, occasionally, for ~$2,000. By June: several could, reliably, for ~$20. Over the next year, we will likely enter a period where AI's offensive capability outpaces…
- likes: 70
- replies: 9
- bookmarks: 25
- reposts: 11
PG Pat Grady@gradypban excellent, opinionated state of the union on AI security
- likes: 52
- replies: 4
- bookmarks: 53
- reposts: 4
