DEV Community

Honey Kiran
Honey Kiran

Posted on Originally published at honeykiran.substack.com on

Two AI Tools Got Hit in One Week. Check Your Agent's Logs.

Two AI tools. Two hits. One week apart. One leaked stored API keys. The other may have exposed agent decision logs. My agents work the 3AM shift so nobody else has to. Their logs remember everything.

Hit one: the keys walked out

OpenCode leaked stored third-party API keys. About 1.4% of OpenCode Console users were affected — the keys sat in their workspaces — and some of those users have reported charges of up to $1,700 run up on keys that were never meant to leave the machine. That figure is user-reported, not vendor-confirmed.

The keys were not stolen from a vault. They were sitting in the tool's own stored data — written down by the tool, for the tool.

Hit two: the diary may be open

Warp disclosed that a vendor in its infrastructure was exploited. What may be exposed: agent decision logs and telemetry — the record of what the agent chose, and why.

Warp's CEO put the affected share at roughly 1%, as a preliminary estimate. What is not clarified yet: whether code, prompts or credentials are in that data. Read that sentence again. The company itself cannot yet tell you the diary is clean.

What your agent writes down

This part is general — true of agent tools as a class, not a claim about any one vendor's logs:

  • Your prompts — including the ones you pasted at 2AM and forgot

  • File paths — a map of your machine and your projects

  • Command output — environment variables ride along more often than anyone admits

  • Pasted config — the YAML you dropped in "just to debug this"

  • Tokens — echoed back in errors, headers and debug lines

Your agent takes notes like a topper before exams. It just doesn't know which notes are secret.

Four checks before tonight's shift

  1. Scan the logs, not just the repo. Run a secret scanner over local agent session and log folders.

  2. Read the sharing settings. Telemetry, session sync, history sharing: turn off what you don't need.

  3. Rotate what showed up. Any key that appeared in a session gets rotated.

  4. Shrink the blast radius. Scoped, short-lived keys with spend caps. "Never expires" is a default, not a plan.

Fence the agent

None of this needs new tooling. Anthropic open-sourced its sandbox-runtime back in October 2025 — a fenced space where agent commands run without touching the rest of the machine. The pattern is a year old. It is shelf stock, not news.

The fence is the direction. The diary stays. What changes is whose hands can reach it.

Your agent's history is either your best audit trail or someone else's payday. It depends on four checks.

When did you last read what your agent writes down?


Disclaimer

  • This post describes my personal experience with my own systems and accounts, in my own words. It is a first-person account of what I observed and what I do in my own work — not a statement of any company’s intentions, policies, or internal decisions. Where I do not know something, the post says so.

  • I am not affiliated with, endorsed by, or speaking for any company, product, or vendor named in this post. All product and company names are trademarks of their respective owners and are mentioned only to describe the experience accurately.

  • Any error messages, documentation excerpts, or figures quoted here are reproduced from my own records or from the public sources linked in the post, believed accurate at the time of writing. Product behavior changes over time; verify against current official documentation before relying on any detail.

  • Nothing in this post is professional, financial, legal, or security advice for your specific situation. Examples are illustrative — validate in your own environment before running anything in production.

  • Corrections are welcome. If any company named here publishes an official statement, or if any detail is wrong, I will update this post and note the correction. The goal is an accurate public record, not a verdict.


© 2026 Honey Kiran. All rights reserved.

Honey Kiran · DevOps & Platform Engineering

AI agents transforming DevOps, infra, platform, security & SRE. The shift nobody works anymore: deep dives, 3AM Files, and OpsBites.

Examples are illustrative — validate in your own environment before running in production.

Top comments (0)