Ridge Security's 96-test pen-test benchmark had Claude Opus 4.6 reach 63% coverage at $217 a run, against 52% for Gemini 3 Flash at about $5.42. Ridge argues tooling matters more than the model, yet its published runs held the tooling fixed and varied the model.
Reality
- Evidence35
- Adoption
- Insufficient
- Hype gap+40
- Incentives70
- Confidence40
A year on from the CISA and NSA memory-safety report, the adoption evidence is still provider roadmaps and practitioner judgement about C and C++ code that nobody proposes to rewrite wholesale.
Reality
- Evidence45
- Adoption35
- Hype gap+20
- Incentives70
- Confidence50
Echo picked up the assets after Minimus decided to close, so this time the customers have somewhere to land. Anyone pinning a base image to one publisher is buying that publisher's rebuild queue too.
Reality
- Evidence40
- Adoption
- Insufficient
- Hype gap+32
- Incentives78
- Confidence42
Oasis Security says CVE-2026-65105 lets one malicious page reach the Ollama instance behind a NemoClaw agent and rewrite the template every prompt passes through.
Reality
- Evidence52
- Adoption
- Insufficient
- Hype gap+18
- Incentives72
- Confidence48
Attackers moved upstream into the project's own repositories and release workflows, so the attestations checked out. Publisher reputation no longer tells you a build is clean.
Reality
- Evidence42
- Adoption28
- Hype gap+22
- Incentives82
- Confidence40