Loading vulnerability…
CVE-2026-48168 is a critical vulnerability in PraisonAI versions prior to 4.6.40 affecting the bundled Claude GitHub Actions workflow. The workflow constructs a Bash run step using a pull request branch name that is attacker-controlled, without proper quoting or validation, creating a command injection condition. The issue is compounded by insufficient authorization checks in the workflow logic, which permits any user to trigger the job by posting an @claude comment rather than restricting execution to trusted collaborators. An external contributor can submit a pull request from a fork with a crafted branch name containing shell metacharacters and then trigger the workflow, causing arbitrary shell commands to execute on the GitHub Actions runner. Because the workflow executes with elevated CI/CD privileges, the initial injection can be chained into later privileged steps.
Mallory correlates every CVE against your assets, your vendors, and active adversary campaigns. Know which vulnerabilities matter for you, not just which ones are loud.
What it means. What to do now. Patch path, mitigations, and the assume-compromise checklist.
What an attacker gets, and what they’ve been doing with it.
If you can’t patch tonight, do this now.
Patch, then assume compromise.
No public exploits tracked yet. Mallory keeps watching.
No public exploit code observed for this vulnerability.
6 sources tracked across advisories, community write-ups, and news. New activity surfaces here as Mallory finds it.
Query your assets running an affected version, and investigate the blast radius.
Every observed campaign linking this CVE to a named adversary.
Malware families riding this exploit, with evidence and IOCs.
YARA, Sigma, Snort, and vendor rules, auto-deployed to your SIEM.
Cross-references every affected SKU, including bundled OEM variants.
Community discussion across Reddit, Mastodon, and other social sources.