CVE-2026-54769 is a critical sandbox escape vulnerability in Langroid affecting versions prior to 0.65.2. The flaw is present in the framework’s TableChatAgent and VectorStore capabilities when LLM-generated tool messages are evaluated with full_eval enabled. The vulnerable implementation attempts to constrain Python eval() by passing an empty locals dictionary, but fails to explicitly remove builtins from the globals mapping. Under Python’s execution model, built-ins are implicitly made available in that situation, allowing evaluated expressions to access dangerous functionality and escape the intended sandbox. Because TableChatAgent’s evaluation path executes external LLM-produced expressions natively, an attacker able to supply crafted prompt payloads can trigger arbitrary code execution on the host.
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.
Products and vendors Mallory has correlated with this vulnerability. Open in Mallory to drill down to specific CPE configurations and version ranges.
Vendor-confirmed product mapping. Mallory continuously reconciles this list against your asset inventory.
7 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.