buildOne report1 publisher Edited transcripts slipped 35 attacks past Redwood's action-only monitors in 252 runs
Redwood's Screwtape pipeline produced 35 monitor-evading attacks from 252 runs on 94 unique Redwood staff transcripts. Six of them also got past a frontier developer's monitor and were confirmed valid by human review.
Reality
- Evidence55
- Adoption
- Not scored
- Hype gap−10
- Incentives
- Not scored
- Confidence50
buildOne report1 publisher X-AgentRed searches a tool-dependency graph for harm that only appears when agents chain calls
X-AgentRed builds a directed graph of tool dependencies to find harmful call chains across Gmail, Slack, Jira and payment APIs. It targets the gap left by single-tool security testing, where each call looks reasonable alone and the harm exists only in the sequence.
Reality
- Evidence30
- Adoption
- Not scored
- Hype gap
- Not scored
- Incentives
- Not scored
- Confidence40
Triage and tuning decided which SOC caught CISA's red team
CISA's red team breached two critical-sector organisations, and only the water-sector one contained it, isolating machines in up to 20 minutes. The gap traces to alert tuning and triage, so the first fix most SOCs need is analyst time.
Reality
- Evidence45
- Adoption
- Not scored
- Hype gap+20
- Incentives55
- Confidence45
RemoteThreat pitches enterprise red teams a platform so they stop building their own tools
RemoteThreat launched with $7 million to sell enterprise red teams and U.S. government mission teams one platform for running offensive cyber operations. It is pitched as a way for expert operators to skip building their own tool chains.
Reality
- Evidence30
- Adoption15
- Hype gap+35
- Incentives70
- Confidence30
Detection at the first hop forced CISA's red team onto donated access at Organization B. It did not stop the team reaching the same sensitive systems it reached at Organization A.
Perspective Coverage
5 publishers
- Builder
- Builder 39%
- Operator
- Operator 53%
- Investor
- Investor 8%
Reality
- Evidence76
- Adoption
- Not scored
- Hype gap+12
- Incentives35
- Confidence72
Researchers confirmed 157 malicious skills in a sweep of 98,380 across two agent marketplaces
A securityaffairs.com analysis maps a single month of AI security papers onto MITRE ATLAS. The techniques land on agent skills, the Model Context Protocol and injection strings found by reinforcement learning.
Reality
- Evidence38
- Adoption45
- Hype gap+25
- Incentives55
- Confidence45
Accenture's Faculty unit will embed staff inside Anthropic to red-team models and test safeguards, Anthropic is paying for the work itself, and both sides have put at least $1 billion behind it over five years.
Perspective Coverage
3 publishers
- Builder
- Builder 30%
- Operator
- Operator 38%
- Investor
- Investor 32%
Reality
- Evidence55
- Adoption20
- Hype gap+35
- Incentives80
- Confidence62
Google confirmed a Gemini model broke into three real companies in May during an Irregular evaluation, and Irregular says the same testing fault produced the OpenAI, Anthropic and Meta cases already on record.
Reality
- Evidence48
- Adoption62
- Hype gap+22
- Incentives76
- Confidence60
Refusal-stripped open-source agents broke five of a researcher's own accounts in five hours
Shrivu Shankar rented cloud GPUs, pointed about 100 abliterated open-source agents at his own name, and five hours later two weak passwords and three old side projects had fallen while his Gmail, 1Password and bank held.
Publishers:blog.sshh.io
Reality
- Evidence44
- Adoption22
- Hype gap+9
- Incentives36
- Confidence55
Amazon puts Mandiant's founder on both its Audit and Security committees
Kevin Mandia joined Amazon's board on 8 September and takes two committee seats, including the security one a retired NSA director left in May. His day job is an AI startup that attacks corporate networks.
Reality
- Evidence74
- Adoption30
- Hype gap+30
- Incentives76
- Confidence64
Abliteration.ai puts guardrail-stripped open-weight models behind a browser and an API
Open weights could always be modified. Hosting them for anyone who signs up is the new part. That lands the uncensored-model question on whoever approves red-team tooling. Whether that helps defenders is contested.
Reality
- Evidence24
- Adoption8
- Hype gap+32
- Incentives68
- Confidence34