High SA-26-727 vLLM: Request-controlled mm_processor_kwargs.code_revision allows remote code execution

Know what can go wrong with AI before it does.

Sourced advisories. Plain-language scorecards. Free tools for small teams.

Securaity tracks security and safety problems in AI software and services, rates how serious they are, and says in plain language who is affected and what to do.

Every fact dated and linked to its source.

LATEST ADVISORIES

See all

Securaity in numbers

Advisories published
727
Products tracked
30
Last update

How Securaity works

  1. 01

    Collect

    Every day we pull new vulnerability records for popular AI software from public databases.

  2. 02

    Verify

    Each record is matched to a named product and checked against its original source.

  3. 03

    Rate severity

    Severity follows the published CVSS score, from Critical down to Info.

  4. 04

    Explain what to do

    Each advisory names the affected product and links to the source with fix details.

  5. 05

    Keep it current

    Records are checked daily, and changes at the source update the advisory.

What every page promises

  • Sourced

    Every advisory links to the public record it comes from.

  • Dated

    Every advisory shows when we published it and when the source last changed.

  • Plain-language

    Short, clear pages for the people who have to decide.

  • Free

    Every page is free to read, with no sign-up.

Explore

  • AI tool safety profiles

    What popular AI tools do with your data, and the settings to change first.

  • AI Safety for Non-Profits

    A free course in ten short lessons for small teams that use AI.

  • Guides

    Frameworks, policies and attacks explained in plain language.

  • Glossary

    Short definitions of the AI security terms you will meet.

  • Blog

    What the advisories show, and how to put them to work.

  • Vulnerabilities by product

    Advisory counts and an RSS feed for each AI product we track.