AI article
Coding Agents Pass Up to 91% of Tasks in a New Benchmark. At Most 25% Are Also Secure
Community description: LSCBench gave six coding agents 450 tasks that never mention security. Up to 91.3% of tasks got working code. At most 25.1% got secure code.
Dev.to | Oct 14, 2026 | Gabriel Anhaia
Automated excerpt
A coding agent rewrote the missing file at the heart of a Java configuration importer. All four security tests failed, because the default SnakeYAML constructor it used "honors every !! class tag" in the uploaded YAML (the paper, Section 5. 4 and Appendix F. 2). The project page breaks down how often runs ended both working and secure, all six models combined.
Selected automatically from source text; not independently written or fact-checked. Read the original for full context.