Venice AI hits unicorn status while already profitable

Tenet Security disclosed a vulnerability class called agentjacking that allows attackers to inject malicious instructions into error data from services like Sentry, which AI coding agents then execute with full developer privileges. Testing achieved an 85% success rate across 100-plus targets, and 2,388 organizations were found with publicly exposed Sentry credentials vulnerable to this attack. The flaw bypasses all traditional security controls because every step in the attack chain is technically authorized.

Meta has implemented strict internal guidelines limiting how its engineers can use Anthropic's Claude and OpenAI's Codex, citing concerns that outputs from these external AI tools could contaminate Meta's own training data. An internal memo instructed teams to pause certain tasks using these models to avoid potential escalations with partner companies. The move reflects Meta's broader effort to reduce dependence on expensive third-party AI coding applications while building internal alternatives.

Palantir has launched a new intelligent engine using NVIDIA Nemotron open models to serve U.S. government agencies in air-gapped, secure environments. The system allows federal agencies to build custom AI models on their own infrastructure, train on proprietary data, and retain full ownership of model weights. This deployment combines Nemotron's open models with Palantir's Sovereign AI Operating System to enable trusted, auditable AI in sensitive government operations.

Prosecutors in the trial of Jonathan Rinderknecht, charged with arson for setting a New Year's Day 2025 fire that became one of LA's deadliest wildfires, used ChatGPT logs as evidence. The logs showed Rinderknecht had asked the chatbot to generate images of fire, inquired about personal anger, and questioned whether someone could be blamed for a fire they did not physically light. The case marks a notable instance of AI conversation records being introduced in a criminal prosecution.