Ground Truth.
AI, checked against the source.

News · 2026-11-14

Codex Security adds resumable standard scans

OpenAI’s Codex Security 0.10.0 adds resumable standard scans and estimated budgets for bulk scans. The changes address an operational problem in AI-assisted security review: analysis must survive interruptions and produce a durable report before a team can act on it.

Key facts

OpenAI calls the project “Codex Security.” Its repository describes command-line, software development kit and automation workflows for investigating code, validating findings, generating reports and reviewing fixes. Some cybersecurity requests and protected findings require Trusted Access for Cyber, so availability of the public client does not imply identical access to every workflow.

A scan is better understood as an investigation than as a single yes-or-no test. An agent inspects code, follows possible paths, gathers evidence, and may try to validate a suspected issue. A useful result needs both a finding and enough context for a human to decide whether it matters and how to fix it.

The new resume capability resembles saving progress during a long document review. If a machine stops halfway through, the reviewer can continue from a checkpoint instead of repeating every step. The distinction between a recoverable scan and a successful final report remains important: checkpoints and artifacts have to be written correctly, and the resumed run still needs review.

A user issue describes analysis that generated findings and coverage but ended with a scan-identifier mismatch during saving. A separate Deep Scan report describes a refusal after preflight and a lack of a supported in-place resume path in the reported version. These reports document possible failure modes; they do not establish current failure rates or prove that the new standard-scan feature resolves every scenario.

Budget estimates also need careful interpretation. Work already in flight can exceed the reported estimate, according to the release notes. A team operating scans at scale should plan for that uncertainty and inspect both coverage and the saved report rather than treating a completed process as a complete security guarantee.

The supplied Trending capture recorded 599 stars for the day, a visibility signal rather than an installation count. No independent measurement of vulnerability detection effectiveness was verified in this research pass. The prompt injection lesson offers relevant background on why any agent processing untrusted material needs explicit boundaries, even though the issue reports above concern persistence and access rather than a demonstrated injection attack.


Primary source, verified: read the paper →

Key questions

What can resume in version 0.10.0?

The release adds resume support for standard scans; earlier Deep Scan reports should not be treated as proof of the same recovery behavior.

Is the estimated bulk budget a hard cap?

No. The release notes warn that in-flight work can exceed the estimate.
Cite this

APA

Ground Truth. (2026, November 14). Codex Security adds resumable standard scans. Ground Truth. https://groundtruth.day/news/codex-security-resumable-scans.html

BibTeX

@misc{groundtruth:codex-security-resumable-scans,
  title  = {Codex Security adds resumable standard scans},
  author = {{Ground Truth}},
  year   = {2026},
  month  = {nov},
  url    = {https://groundtruth.day/news/codex-security-resumable-scans.html}
}

Topics: cybersecurity · ai-security · coding-agents · security-scanning

Comments are replies to this story on Bluesky — reply with any Bluesky account to join in.