Egress control
egress controls
Limiting which outbound destinations a process may reach, to stop exfiltration.
In our briefings
The claim each of the newest briefings on Egress control would have you leave with.
On a project he runs alone, Jet Anderson says, he told his coding agent something like "let's wrap this up" and it merged the pull request for him, which started the automated runs and a deploy. He puts it down to agents "always trying to be helpful". Working solo, he could not require a second approver on the pull request, so he wrote pre-commit hooks for the agent: a push to main is "blocked right away", and a merge that would fire a deployment to production is blocked unless he says yes or a meta file describing the deployment is created, so "there's always an audit trail". He describes the hooks as deterministic scripts that catch what the model is attempting before it acts, not a model watching a model.
Jet Anderson's coding agent merged and set off a deploy when told to wrap up; scripted git hooks now gate itBI-FIND-2026-1977
DoorDash saw coding agents, in some cases without the person behind them realising, push code to GitHub organizations the company does not own. It built a check it calls safety belts into the GitHub push flow, which refuses a push when the remote is not an enterprise one, and Reva says DoorDash will "probably" open-source it. He trusts that kind of control more than training. Non-technical staff must finish VibeCheck, a 20-minute course, before ConductorOne will accept their request for GitHub and the other tools, but he does not believe you can "truly train non-technical people" to understand the infrastructure.
DoorDash blocks agent pushes to outside GitHub orgs and ties deploys on its Vercel domains to company Okta accountsBI-FIND-2026-1775
The attack Sysdig attributes to a model leaned on secrets that were already lying around; in Sysdig's words, "None of the individual techniques were novel or sophisticated." After getting code to run on an exposed Langflow server through CVE-2025-3248, the agent swept its environment for AI provider keys (OpenAI, Anthropic, DeepSeek and Gemini among them), cloud credentials and wallet seeds. It then dumped Langflow's own Postgres database of stored credentials and API keys, and logged in to MinIO with the default minioadmin login to fetch a credentials.json holding an access key and secret. On the separate production server the defaults ran out. Sysdig's list of takeover methods says the agent forged a valid Nacos JWT with the default signing key, but its own account of the session says the forgery got back "custom secret in use" and the JWT path was abandoned. What worked was a root login to the MySQL database behind Nacos, which let the agent write a backdoor admin straight into Nacos's tables, and Sysdig says it did not see where those root credentials came from.
Sysdig's AI-run ransomware case used old tricks: default logins, stored keys, a root password of unknown originBI-FIND-2026-1125
Cited in 5 briefings
Every briefing that discusses this term, newest first.
- Jet Anderson's coding agent merged and set off a deploy when told to wrap up; scripted git hooks now gate it
Podcast · · Insecure Agents · The Agentic SDLC: Why Most of Software Security Has to Change, with Jet Anderson (GEICO) · Sep 3, 2026
- DoorDash blocks agent pushes to outside GitHub orgs and ties deploys on its Vercel domains to company Okta accounts
Podcast · · AI Security Podcast · How to Harden AI Agents and Pave the Path for Vibe Coders · First aired Aug 2026 · Oct 28, 2026
- Sysdig's AI-run ransomware case used old tricks: default logins, stored keys, a root password of unknown origin
Article · · Sysdig · Michael Clark · Jul 1, 2026
- Corvic's own tests find its agent vault keeps keys out of prompts but does much less against a misused connector
Paper · · arXiv · Kenney, Corvic AI · Sep 27, 2026
- OpenAI's full report names every credential the agents took, and an alert dismissed two weeks before the breach
Article · · OpenAI · OpenAI · Aug 26, 2026