English NewsSunday, 15 November 2026
Latest
Arrest warrant issued after absence from coastal statue caseTwo respondents summoned over alleged breach of online-posting orderLandslide alerts cover parts of seven districts until November 12 eveningUnregistered shop signs removed along Adampan main roadCocaine case involving three defendants set for January 25 pre-trial conferenceMirissa guide says viral boat video showed play between friendsWhat could a change of government in Israel mean for relations with India?Sri Lanka keeps electricity tariffs unchanged for final quarterFive-working-day business registration pilot begins in selected Sri Lanka divisionsDrone taxi project takes flight at Colombo Port CityPooja Hegde draws crowds at Colombo fashion-store openingMinistry explains private fuel price flexibility and plans 72 CPC stations
AI

Independent control assessment gives leading AI labs weak containment grades

2026-08-19

published reporting control assessment gives leading AI labs weak containment grades
Photo: Mona Hassan Abo-Abda (CC BY-SA 4.0)

Guidelight’s August control assessment, current through 18 August, graded Anthropic and OpenAI C+ (2.50), Google D+ (1.50), xAI D− (0.83) and Meta F (0.67). It scored public evidence for logging, monitor effectiveness, gated actions, circuit breaking, third-party review and containment planning.

Why it matters: The report moves AI control from broad principles to inspectable operating practices, but it relies on public disclosures and is neither a regulator audit nor peer-reviewed research.

Business/technical/strategic impact: Boards can use the six practices as a due-diligence checklist. Scores should prompt evidence requests, not be treated as definitive measures of internal controls.

More AI technology news

More in AI

Weak containment scores for leading AI labs

Weak containment scores for leading AI labs

Guidelight assessment says frontier labs still lack mature AI-control practices

Guidelight assessment says frontier labs still lack mature AI-control practices

No mature AI-control practices at frontier labs, Guidelight assessment finds

No mature AI-control practices at frontier labs, Guidelight assessment finds

Anthropic reportedly agrees a US$45 billion, six-year Nscale compute contract

Anthropic reportedly agrees a US$45 billion, six-year Nscale compute contract

Google gives publishers a “Preferred Sources” button

Google gives publishers a “Preferred Sources” button