Anthropic advarer: Z.ai sin GLM-5.3 kan lage cyberangrep uten sikkerhetstiltak

16 kilder
  • Anthropic hevder at Zhipu AI sin GLM-5.3 autonomt kan lage fungerende cyberangrep på nivå med deres egne toppmodeller, men at den er lansert uten effektive sikkerhetsmekanismer.
  • Forskere reduserte modellens avvisningsrate fra over 90 % til rundt 3 % ved hjelp av en standardteknikk som kostet omtrent 4 400 dollar i datakraft, ifølge Anthropic.
  • NIST har kalt GLM-5.3 "den mest cyber-kapable åpen-vekt-modellen som er utgitt til dags dato", selv om enkelte analytikere stiller spørsmål ved om den endrer trusselbildet nevneverdig.
Kilder (16)
  1. 1 GLM-5.3 and the spread of advanced cyber capabilities www.anthropic.com
  2. 2 Anthropic Warns GLM-5.3 Builds Exploits Like Mythos | Trending Topics www.trendingtopics.eu
  3. 3 CAISI's Assessment of Z.ai's GLM-5.3 Cyber Capabilities | NIST www.nist.gov
  4. 4 Anthropic Warns Z.ai's GLM‑5.3 Vulnerable to Cyber Misuse www.marketscreener.com
  5. 5 CAISI Assessment of Z.ai's GLM-5.2 | NIST www.nist.gov
  6. 6 Zhipu’s GLM-5.3: Coding AI with Cybersecurity Capabilities www.whats-ai.com
  7. 7 GLM-5.3 and the Spread of Advanced Cyber Capabilities \ Anthropic www.reddit.com
  8. 8 Center for AI Standards and Innovation (CAISI) | NIST www.nist.gov
  9. 9 Anthropic: Zhipu's GLM-5.3 Matches Claude on Autonomous ... aiweekly.co
  10. 10 GLM-5.3 and the spread of advanced cyber capabilities | Hacker News news.ycombinator.com
  11. 11 CAISI's Open-Weight Model Evaluation Cadence - CASRAI casrai.org
  12. 12 Chinese AI Giant Zhipu Unveils GLM-5.3, Edging Past Western ... www.chinatechnews.com
  13. 13 GLM-5.3: Frontier Coding with Emergent Cyber Capabilities - Z.ai z.ai
  14. 14 Cyber AI Assessments Need Expiry Dates and Context - LinkedIn www.linkedin.com
  15. 15 China's GLM-5.3 Leads a Cyber Benchmark and Its… | AI in Asia aiinasia.com
  16. 16 GLM-5.3 Cyber Performance www.nist.gov