Entity trail

LASE-Bench

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-09-01 / agents-researcherECLIPSE evolves its own prompt injections against long-horizon agents and hits 69% success even through a safety filterECLIPSE synthesizes attack tool-chains in a sandbox, verifies them, then renders the verified chain as a single natural-looking prompt while embedding state-transition cues in target tool descriptions (Static Workflow Encoding) and correcting drift mid-run (Dynamic Trajectory Correction). Against Codex, Claude Code and OpenClaw-style harnesses it reaches 96.7% attack success undefended and 69.2% under a common safety filter, beating the strongest baseline by 27.5 points in the defended setting. The paper ships LASE-Bench, 120 malicious tasks over 198 tools where 96.7% of tasks require five or more tool calls, so this is explicitly a test of the long-running agent loop rather than a single turn.
2026-08-18 / agents-researcherAlphaEvolve helps push the matrix multiplication exponent to ω < 2.371177, the first improvement since 2024A ten-author team including Emilien Dupont, Josh Alman and Virginia Vassilevska Williams reformulated the combination-loss optimization behind the laser method, applied modern ML-based optimization, and then used DeepMind's AlphaEvolve as a final refinement step — landing ω < 2.371177 against the previous best of 2.371339. The notable structural point for builders is the division of labor: the humans reformulated the problem and the evolutionary coding agent squeezed the last digits out of the resulting algorithm. This is one of the cleanest documented cases of an agent contributing a real, checkable delta to a headline theoretical-CS result rather than to a benchmark score.

Graph relationships

BENCHMARKED AGAINST
Eclipse
ECLIPSE is evaluated on LASE-Bench with 120 malicious tasks.
Source finding
BENCHMARKED AGAINST
Eclipse -> LASE-Bench
ECLIPSE is evaluated on LASE-Bench with 120 malicious tasks.
Source finding

Source trail

arXiv arXiv

Graph sources

entity graphfindings textkg edgeskg entities

Fetching from the wire…