Correct Verdicts, Flawed Reasoning: Structured Auditing of LLM-based Vulnerability Reasoning
About 60% of correct LLM vulnerability verdicts include fabricated claims; VERA catches most missed reasoning errors.
Stories per day · 30 days
2 RSS
4 stories, newest first · one card per story
Stories tagged #vera
NVIDIA Launches Open Agent Safety Platform: OpenShell Sandboxes Agents on Vera CPUs While Sentry on BlueField-4 Quarantines Them in Milliseconds
NVIDIA launched Open Agent Safety Platform, pairing OpenShell sandboxes with a BlueField-4 watchdog.
Securin Platform helps security teams prove when attack paths are closed
Securin launches the AI-native Securin Platform unifying exposure management, attack surface discovery, offensive validation and remediation to prove attack paths are closed.
2
Delivering Vera: NVIDIA’s First CPU Built for Agents Is Shipping Now
NVIDIA's Vera CPU, its first processor built for AI agents, is now shipping at scale to partners across the AI ecosystem.