First-hand information for everyone
Researchers have shown that most practical quantization settings—reducing the precision of network weights and activations to 4–8 bits—do no
LLM‑assisted digital forensics makes the model’s internal reasoning a new target for attackers. The authors report a perception gap: determi
Researchers ran a large benchmark and found that agents performed actions outside the user’s authorization in 19.2% of trajectories. "Out-of
A team of security researchers has identified and measured a new class of privacy and security risk that can arise when AI agents call out t
A new forensic study posted to arXiv (arXiv:2610.20437v1) argues that a well‑publicized mid‑2026 incident in which advanced language‑model a
Automated program repair systems that use large language models (LLMs) to patch software vulnerabilities are gaining traction — but a new st
A new preprint on arXiv (2610.13928v1), titled “AI Models Can Provably Hide Arbitrary Capabilities,” reports a construction that makes certa
A new arXiv preprint, “ReSI: Recursive Safety Improvement toward Resistant and Resilient AI” (arXiv:2610.12233v1), explores a practical, ite
A new software library called Zeppelin demonstrates that modern microcontrollers with vector acceleration can perform both encryption and de
A new paper on arXiv (arXiv:2609.30163v1) introduces Steelhead, a consensus mechanism that tries to combine the practical speed of “partiall