Tag
User @Saccc_c shares the experience of using the newly released GLM 5.3 to create a landing page website, emphasizing its 50% improvement in coding capability, leading in open-source models, and being on par with Kimi K3 in front-end coding.
CSIS report urges the US to combine AI-enabled monitoring and rapid patching with systematic reviews of frontier AI models before and after release to strengthen cybersecurity.
This survey examines LLM unlearning methods for cyber defense, introducing a three-level framework to distinguish behavioral suppression, representation-level attenuation, and true forgetting, and analyzing gradient-based, influence-based, and localized editing approaches.
Article argues that export controls on AI models like Claude Fable 5 harm US cybersecurity by banning the ability to fix code vulnerabilities, which is essential for defensive security. The controls are based on a misunderstanding of AI capabilities.
A controlled study of compound LLM agent design in an adversarial POMDP (CybORG CAGE-2), systematically varying context, reasoning, and hierarchy across five model families. Key findings: programmatic state abstraction yields large returns per token, hierarchy without deliberation tools achieves best absolute performance, and context engineering is more cost-effective than deeper reasoning.
A previously stealth cyber defense research lab emerges to develop AI-driven superhuman cyber defenders for Western security advantage.
OpenAI outlines comprehensive security measures on the path to AGI, including AI-powered cyber defense, continuous adversarial red teaming with SpecterOps, and security frameworks for emerging AI agents like Operator. The company emphasizes proactive threat detection, industry collaboration, and security integration into infrastructure and models.
Anthropic's Project Glasswing, using Claude Mythos Preview, has found over ten thousand high- or critical-severity vulnerabilities in critical software, with partners like Cloudflare reporting a tenfold increase in bug-finding rate.