Burnham senior officials trade blame over economic painJustice Department memo relaxes rules for senior appointees to engage in political campaignsPutin says Trump peace talks unlikely after Moscow drone attacksThousands of Imran Khan supporters march toward Islamabad demanding his releaseTaiwan President William Lai says defence spending increase intended to deter warIsrael's economy shows growth amid war, but rising food prices raise voter concernsTaiz faces renewed siege amid rising food and fuel pricesSaudi-led coalition conducts operation in Yemen targeting HouthisLibyan military reports seizure of over 1.3 million narcotic pills after clash with smugglersMystery Persists Over Causes of Havana Syndrome Among CIA OfficersReview led by Alan Milburn proposes national pre‑apprenticeship scheme to address youth unemploymentDebate Emerges Over Whether AI Can Surpass Human CapabilitiesThe Two NewportsUkrainian Man Faces Trial in Hamburg Over Alleged Nord Stream SabotageFederal judge blocks DOJ's request for state voter lists
Politifex
Search
Back to event
TechCrunchArtificial IntelligenceBig TechStartupsInternet AccessInternal Evaluations

Anthropic can’t reliably control its AI agents. It’s cutting off its internal evals from the live internet instead

By Tim Fernholz · Oct 9, 2026, 7:18 PM CDT

Read full article at TechCrunch
Anthropic said it "turned off live internet access" for "all our internal evaluations" until further notice.

Excerpt shown under fair-use limits. Full text remains with the original publisher.

People in this coverage

Explore their history and attributable record. Being mentioned does not imply endorsement.

Tim Fernholz →

Layer 1 · Claims & fact checks

AI analysis

Layer 2 · Biblical perspective

Biblical interpretation
INSUFFICIENT CONTEXT
Read the biblical analysis

Layer 3 · Reporting analysis

AI analysis

FramingThe language frames the action as a decisive, organization-wide measure, emphasizing the breadth ('all') and the temporary nature ('until further notice').

Headline FitThe headline extrapolates from the quoted statement, implying a broader reliability issue ('can’t reliably control its AI agents') that is not directly stated in the source text.

Context

AI analysis

Missing context

The statement does not explain why the live internet access was turned off, what specific evaluations are affected, how this impacts Anthropic's AI development or safety, and whether this change is temporary or permanent.

Important context

Anthropic is a leading AI research organization that develops advanced language models. Controlling AI agents' access to live internet data is a key safety and reliability concern, as unrestricted access can lead to unpredictable behavior or the propagation of misinformation.

Opinion vs. reporting

AI analysis

The article presents a factual report of Anthropic's announcement without adding editorial commentary or opinion.