RepoJournal
Agent Security Agent Security
37 wires and counting

$ follow Agent Security

Keep up with Agent Security in about 3 minutes: what actually shipped — the commits, pull requests, releases, and security advisories that matter.

or

fair warning: these emails are deeply technical. diffs, version numbers, CVEs, benchmark deltas. if that's not your idea of a good read, this isn't your newsletter.

Folds into your digest — weekly by default, monthly if you prefer. Unsubscribe in one click.

$ status

wire 2026-11-07
stories 41

© 2026 RepoJournal Home Showcase How it works Privacy

$ the-wire · showcase

Injection rail now blocks when redaction is incomplete

By RepoJournal · Filed · About Agent Security · Composed from the cited sources · methodology

NeMo Guardrails hardened its injection detection so unsafe output gets blocked instead of passed through, and swept three dependency advisories in the same window.

fix(library): prevent detection bypasses and unsafe omission in injection_detection NVIDIA-NeMo/Guardrails

by Pouyanpi

The injection_detection rail closes occurrence-order bypasses in the Python, Jinja, and XSS rules, removes matched byte ranges safely, and rechecks the omitted output against every configured rule, blocking unsafe results or processing errors rather than forwarding them. Both Colang dialects and IORails are covered, and the PR reports a differential check over 157,878 synthetic token sequences ...

fix(library): block output when omission is incomplete in injection detection rail (#2558) NVIDIA-NeMo/Guardrails

by Pouyan

A follow-up to the same rail now blocks output when the omission is incomplete, closing the gap where a partially redacted response could still reach the caller. If you run an injection detection rail in production, this is the behavior you want before your next upgrade.

fix(deps): upgrade Pillow to 13.0.0 for security advisories (#2581) NVIDIA-NeMo/Guardrails

by Tim Gasser

Pillow moves to 13.0.0 in the Guardrails dependency set to clear security advisories, which matters to operators whose guardrail pipelines decode images from untrusted input.

fix(deps): update urllib3 and setuptools for security advisories (#2565) NVIDIA-NeMo/Guardrails

by Pouyan

urllib3 and setuptools are updated for their own security advisories; both sit under the HTTP stack Guardrails uses, so the fix rides along with any environment rebuild.

chore(deps): bump certifi from 2026.6.17 to 2026.10.18 NVIDIA-NeMo/Guardrails

by dependabot

Also in the queue: dependabot moved certifi from 2026.6.17 to 2026.10.18, and snyk/agent-scan removed no-op agentic fix parameters, neither of which changes how you configure either tool.

Quick answers

What shipped in Agent Security on November 7, 2026?
NeMo Guardrails hardened its injection detection so unsafe output gets blocked instead of passed through, and swept three dependency advisories in the same window. In total, 12 commits, 28 pull requests, and 1 releases landed.
Who contributed to Agent Security on November 7, 2026?
4 developers shipped this update, including Pouyan, Tim Gasser, dependabot, and Hemang Sarkar.
What were the notable Agent Security updates?
fix(library): prevent detection bypasses and unsafe omission in injection_detection, fix(library): block output when omission is incomplete in injection detection rail (#2558), and fix(deps): upgrade Pillow to 13.0.0 for security advisories (#2581).