TRACE watermarks LLM-agent trajectories with two keyed channels, trying to survive log deletion and rewriting, which is the annoying part.

Can your negotiation bot's concessions, timing, and convergence leak hidden constraints? This paper formalizes that and randomizes around...

Concordium's pitch: verify age/region/authority with ZK proofs and trusted IdPs, not by shuffling passports around the internet.

Cloudflare's meter is now on, with x402 and Polygon's agent-payment stack for per-request web access.

ToolPrivacyBench, a benchmark for purpose-bound privacy in tool-using LLM agents, because apparently prompts needed audits.

Credentials leak from third-party agent skills more often than you'd hope, per a large-scale arXiv study.

Security roundup: package proxies, AI-agent canary benchmarks, and OpenAI's Daybreak/Codex updates, plus the usual cloud mess.

Survey of privacy risks in LLM agents, from retrieval to memory to prompt injection, with IF control as the lone adult in the room.

OCELOT meters privacy leakage over an agent's whole trajectory, using witness-verified declassification instead of output filters.