Reflection AI says its first model, Beam, performs at the level of Z.ai's GLM-5.2 while using three to four times less compute than similar open models. Until outside testers report, those are the company's figures, and they count most with buyers who cannot use Chinese models.
Perspective Coverage
12 publishers
- Builder
- Builder 37%
- Operator
- Operator 28%
- Investor
- Investor 35%
Reality
- Evidence35
- Adoption10
- Hype gap+45
- Incentives75
- Confidence60
Britain's pre-launch AI safety testing plan lapsed with the government that drafted it
Ministers explored forcing the largest AI companies to submit products for safety testing before launch. The plans lapsed with Starmer's premiership, and the department that was writing them has since been abolished.
Reality
- Evidence58
- Adoption30
- Hype gap+18
- Incentives65
- Confidence55
King Charles brings the leaders of Nvidia, Google DeepMind, OpenAI and Anthropic to Ayrshire this week to weigh a charter of principles the Ditchley Foundation has drafted for them. The push to slow down came from inside the labs.
Perspective Coverage
4 publishers
- Builder
- Builder 21%
- Operator
- Operator 43%
- Investor
- Investor 36%
Reality
- Evidence55
- Adoption18
- Hype gap+30
- Incentives70
- Confidence62
buildOne report1 publisher Three July evaluation runs without standard safeguards gave Claude access to real systems
Anthropic places the three incidents inside cybersecurity evaluations, and reports that the checkpoint behind its deliberately reward-hacking Opus stayed clean on the same tests, which puts training practice in the control surface as well.
Publishers:agentuptime.substack.com
Reality
- Evidence55
- Adoption48
- Hype gap+12
- Incentives70
- Confidence56
buildOne report1 publisher A White House request deleted the public record of federal pre-release model testing
Commerce's AI Standards Center said on May 5 it had access to three frontier models before launch, and the post came down days later at White House request, which is a fair measure of what these voluntary deals are secured by.
Reality
- Evidence26
- Adoption31
- Hype gap+12
- Incentives66
- Confidence34
buildConfirmed2 publishers A Siemens-specific follow-up to the July PLC warning describes internet-wide discovery paired with AI-generated Python tooling that reads and writes ladder logic.
Reality
- Evidence58
- Adoption35
- Hype gap+15
- Incentives40
- Confidence55
The chain of command survives AI; the judgment inside it may not
A Foreign Affairs essay argues the real military AI risk is cognitive: automation bias and deskilling hollowing out the judgment a chain of command assumes it has. Delegation is the exposure.
Reality
- Evidence28
- Adoption
- Not scored
- Hype gap+18
- Incentives
- Not scored
- Confidence34