This release includes 3 security fixes for security teams reviewing exposed deployments.
Topics
+13 more
Affected surfaces
Summary
AI summaryCritical security release fixes arbitrary file write, env variable interpolation bypass, and LLM/API key leakage vulnerabilities.
Full changelog
Security release addressing three advisories reported via GitHub private vulnerability reporting and published 2026-07-07. All users should upgrade from 2.26.5.
Security fixes
- GHSA-2956-977x-2w3r (Critical) — Arbitrary file write via
image.downloadand other file-writing modules. Writes are now confined toFLYTO_SANDBOX_DIRvia the central path guard. - GHSA-hr7p-wg7r-hg9m (High) —
${env.VAR}interpolation bypassed theenv.getdenylist.${env.*}is now deny-by-default, opt-in viaFLYTO_ENV_VAR_ALLOWLIST. - GHSA-qq9q-xgm3-xv9g (High) — Environment-derived LLM/API keys could be sent to a caller-controlled
base_url. Env keys are now only attached to the official endpoint or hosts onFLYTO_TRUSTED_LLM_HOSTS.
Upgrade
pip install --upgrade flyto-core==2.26.6
Reported by @kaimandalic. See CHANGELOG.md for full details.
Security Fixes
- GHSA-2956-977x-2w3r (Critical) — Arbitrary file write via image.download and other modules now confined to FLYTO_SANDBOX_DIR.
- GHSA-hr7p-wg7r-hg9m (High) — ${env.VAR} interpolation bypass fixed; env.* is deny-by-default, opt-in via FLYTO_ENV_VAR_ALLOWLIST.
- GHSA-qq9q-xgm3-xv9g (High) — Environment-derived LLM/API keys now only sent to official endpoint or hosts on FLYTO_TRUSTED_LLM_HOSTS.
Weekly OSS security release digest.
The CVE patches and breaking changes that affected production tools this week. One email, every Sunday.
No spam, unsubscribe anytime.
Share this release
About flytohub/flyto-core
Deterministic execution engine for AI agents with 412 modules across 78 categories (browser, file, Docker, data, crypto, scheduling). Features execution trace, evidence snapshots, replay from any step, and supports both STDIO and Streamable HTTP transport.
Related context
Beta — feedback welcome: [email protected]