Fetching from the wire…
Research2026-08-11 · source-backed
SpecPath (arXiv 2608.09799) holds repository, final contract, verifier, agent system, and execution budget constant, varying only the revision history by which a spec became final. Same end state, different path there. Thirty-five percent failed on at least one contract-equivalent history, and aggregate accuracy across paths looked nearly identical, which means path sensitivity is completely invisible to standard benchmarking. Agents are partly following surface cues from how a request evolved. Evaluate against messy multi-turn requirement histories, not the tidy ticket you'd write in hindsight.
Each link below shares sources, entities, or timing with this story.
SpecPath found 35 of 100 passing implementations broke when only the revision path changed, with aggregate accuracy looking identical across paths. Build your eval set from real multi-turn clarification threads with amendments and reversals. Path sensitivity is invisible to st...
A single-author study models supersession in inherited agent memory: a constraint true when written, since withdrawn by a newer authoritative record. Given a two-record verification budget, agents checked the provenance path in about one episode in five and produced stale-cons...
Cyera signed an LOI to acquire Oasis Security for roughly $1 billion, about $700M in cash (SecurityWeek). Act Security came out of stealth with $60M total, a $20M seed from Team8 and Bessemer plus a $40M Series A led by Notable Capital (SecurityWeek). Hush Security closed a $3...
40+ experts and 31,000+ human hours built 60 project-level research tasks across 11 domains, designed so guidance is progressively withdrawn *within the same project*. Across 18 agent-model configurations: 50.91 with full methodological guidance, 29.10 when only the method is...
Seven models. Five harnesses. Controlled fact-withholding with injected faults. arXiv 2608.16630 is the most operationally direct paper I've read on harness design, and it produces three results that each change what I do this week. One: availability decides outcomes, not dist...
Top of Hacker News today at 424 points and 91 comments: a writeup of driving GPT-5.5 through Codex on a $200 ChatGPT Pro plan, with Claude Pro at $20 acting as an advisor, through more than 1,500 submissions over 14 days to optimize a batched compact-Householder QR kernel on a...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.