Giving Your Agent Your Own API Token: An Infra Engineer's Cautionary Tale
WesternNo2429 · reddit · 2026-11-18
An infrastructure engineer shares on Reddit how lazily handing AI agents his personal API token gave them full access to every system under his identity — indistinguishable in logs, and one poisoned ticket away from a data breach. Reading about this summer's OpenAI agent escapes (which exploited credentials in env vars and unwatched accounts) pushed him to rebuild his approach:
- Agents should have their own identity, not borrow yours, for auditability
- Access should be partial and time-boxed, revoked at the next request
- Anything moving money or changing permissions needs human approval
- Fail closed: if a check can't run, the answer must be no
He built a Rust gateway enforcing "deny unless proven" and asks the community how they handle agent access — personal tokens, service accounts, or gateways.
More from coding & agent
- Dev teases faster, cheaper game creation with his bug-catching VLM stack — OdinLovis · 2026-11-18
- Dev post-trains a VLM to catch game bugs Claude misses: 56% to 81% — OdinLovis · 2026-11-18
- Pipecat open-sources Audience: a tiny classifier so voice agents know when they're being addressed — alexcovo_eth · 2026-11-18
- Weave: Model Writes One Script, Rust Runtime Parallelizes Independent Tool Calls — Business_Rain_1042 · 2026-11-18
- Microsoft open-sources Work IQ Starter Pack to fill demo tenants with realistic data — DanWahlin · 2026-11-18
- GitHub publishes harness-level data comparing Copilot CLI vs Codex and Claude Code — 0xkarasy · 2026-11-18