Anthropic Institute
AI overview
Sign in and the AI will write an overview from our coverage.
Headlines · 3
- AI auditors detect agent misconduct, but high detection targets bring false alarms
Andy Hall’s pilot study replicated and extended METR’s work by testing whether AI models can identify reward hacking and sandbagging in agent logs. At a 5% false-alarm rate, GPT-5 detected nearly 98% of reward-hacking cases, while the auditors caught 71% to 81% of sandbagging cases. The trade-off became more pronounced at demanding thresholds: in a second test using the R-Judge dataset, nearly all models stayed below 10% false alarms when targeting 95% detection, but the best model’s false-alarm rate rose to 15% at 99% detection. Hall concluded that AI auditors are better suited to estimating incident frequency than to reliably catching every incident, while noting that the analysis reflects his personal views. The reward-hacking results rest on only 47 cases, and most sandbagging cases were synthetic.
The Decoder · 🔥 0 - Anthropic’s $44T US GDP scenario is conditional, not a forecast
Anthropic co-founder Jack Clark said an internal economic model puts US GDP at $44 trillion by 2030 under an extreme AI scenario, compared with slightly more than $30 trillion currently. The projection assumes continued technical progress and broad adoption of AI, while a 36Kr analysis noted that Anthropic’s model also produced lower $34.1 trillion and $36.3 trillion scenarios and questioned whether the evidence supports the extreme case. Clark used his IMF speech to call for international rules and tighter safeguards as the policy window narrows.
IT之家 AI · 🔥 0 - Anthropic survey: 91% of staff vs 39% of US adults expect Nobel-level AI by 2030
Anthropic Institute published The AI Expectations Gap on October 16, a survey conducted in August 2026 comparing a nationally representative sample of US adults with Anthropic employees. The two groups broadly agree on what AI can do today, but staff on average expect AI to handle tasks equal to 75% of their current work time by 2030, versus 28% for US workers; 91% of staffers versus 39% of adults expect AI could make a Nobel-level discovery by then. The report also says US adults expect only 42% of technically automatable work to actually use AI by 2030 against 65% for staff, and that the top adoption barrier adults cited was customers preferring to deal with humans. It stresses the figures are expectations, not measured model performance, and that its modelled GDP scenarios (median 5.4% for the public, 17.1% for staff) depend on assumptions.
虎嗅 AI · 🔥 0
Experience and discussion from the community
Share my Anthropic Institute experienceAsk about Anthropic Institute
Nobody has shared their experience with Anthropic Institute yet.