The tool update gap: why agent tools must be re-checked when their code changes
New research shows a tool approved once can turn malicious in an update that prompt-injection defences never see. What it found, its limits and what to change.
Papers, studies and surveys, and what they mean in practice.
New research shows a tool approved once can turn malicious in an update that prompt-injection defences never see. What it found, its limits and what to change.
Research shows agents that flatten history before compaction can record a planted line as the user's own words. What was measured and what to change.
New research finds low attack success rates mostly mean agents ignored injections, and detector rankings don't transfer. How to read these evaluations.
5 min read
Daily coding agent use is the norm, cost controls are mostly blunt limits, and developers switch tools for quality, not price. What leaders can take from it.
A new preprint audits ten agent runtimes and finds approvals that don't bind to what executes. The findings, the proposed fix, and what to ask of your tools.
Researchers hid instructions in AGENTS.md-style rule files and got coding agents to swap real dependencies for attacker packages. What it means for teams.
Terminal-Bench 4.0 is a new task set, not a harder 2.1, and vendor and independent scores already differ. How to read the numbers before they reach a decision.
4 min read
SWE-Bench Pro Verified closes gold-patch leakage channels and refines 102 tasks. Some model scores fall sharply once agents can no longer fetch the solution.
4 min read
A 718-firm study finds AI agents raise code volume and PRs while resolved issues stay flat and review time rises sharply.