Researcher Uses AI to Build a Chrome Exploit and Break Into PostHog Production
Harsh Jaiswal of Hacktron found that PostHog launched Playwright Chromium with the --no-sandbox flag when generating heatmap screenshots. A Hacktron scan flagged it. Rather than report the setting on its own, he set out to show the full impact, working with Claude to find a suitable known V8 (Chrome engine) vulnerability.
The chosen flaw was CVE-2026-7899, which works on x86 systems. After about 36 hours he had remote code execution inside a PostHog production worker container. He says the model's first attempt stalled on leaking a memory detail against production, where an offset differed from his lab, and a newer model version got past it. From the container he could see database credentials, encryption keys and API secrets stored in plaintext environment variables, which meant access to the production database.
The research came to light when evidence in his draft report, saved in a secret GitHub Gist, triggered GitHub secret scanning. That alerted PostHog's security team, who fixed the bug, rotated secrets and credentials, and treated it as an incident. Their advisory is PSA-2026-00001.
Jaiswal also used the exploit to test open-weight AI models. He reports that Opus 4.8 and Grok 4.5 reached full code execution, while Kimi 2.7 and GLM 5.2 got as far as a sandbox escape. DeepSeek was the weakest on this task. His broader point is about patching: PostHog's worker ran Chrome 139.0.7258.5, which had a known V8 flaw. Such gaps were once survivable because turning these bugs into working exploits took rare skills, and he argues AI models are now closing that window.
Why it matters
Teams have treated an outdated browser inside a backend worker as low risk because weaponising a V8 bug needed rare expertise. This case shows that assumption failing: a disabled sandbox plus a known n-day gave production access in about 36 hours with AI help. The plaintext secrets in the container turned a single foothold into database access, so any business running headless browsers for screenshots or rendering is exposed.
What to do
- Remove the --no-sandbox flag from any Playwright or Chromium launch configuration in production.
- Update Chromium and Chrome in worker container images, since version 139.0.7258.5 was exploitable via CVE-2026-7899.
- Move database credentials, encryption keys and API secrets out of plaintext environment variables.
- Rotate any secrets that may have been exposed, and check public Gists and repositories for leaked evidence or credentials.