Agent opsec isn't solved: frontier models are safer but too pricey for every task
craigbalding · x · 2026-11-19
Security researcher Craig Balding notes frontier models hallucinate less and make fewer mistakes, making it tempting to assume agent opsec is becoming a solved problem—fewer API key leaks, better security decisions?
His take: "kind of." The catch is cost—who will pay frontier prices for every agent on every task? Cost constraints mean many deployed agents will run on weaker models, so agent security remains far from solved.
More from Safety
- Simple Unmark adds Confidential mode: browser-verified TEE encryption keeps text away from backend — haltakov · 2026-11-20
- LangChain Launches Tool Gateway for Governed, Auditable Agent Tool Calls — LangChain · 2026-11-20
- Frontier models can unmask authorship from a single text, no tools needed — MohitIyyer · 2026-11-20
- User says OpenAI deactivated his account over light Codex use, appeal gets no human review — hexxthegon · 2026-11-20
- vCenter zero-day CVE-2026-123890 exploited since September, grants admin without credentials — cyb3rops · 2026-11-20
- Meta's AI agent Muse mixes up conversations and exposes users' private data — TechCrunch AI · 2026-11-20