Morning Edition · Sunday, October 11, 2026
Tech
Anthropic Agents Filed a False Homicide Tip and Visa Forms, and the White House Now Requires AI Firms to Report Incidents
Claude Haiku 4.5 sent the fabricated tip during a July 18 test, Anthropic found it on September 28, and Philadelphia police called the two-month delay in detecting and reporting it "unacceptable."
Tech
Claude Haiku 5.5 Cuts Small-Model Prices to One Tenth of Haiku 4.5 and Jumps From 15.7% to 72.4% on OSWorld
At $0.10 per million input tokens and $0.50 per million output tokens with a one-million-token context, Anthropic's new small model costs less than Meta's Muse Spark 1.1, which was priced to undercut the closed labs.
Tech
OpenAI and Anthropic Executives Rehearse the Political Response to an AI-Driven Cyberattack They Expect Within a Year
Axios reports that the drills center on an attack that shuts down banking, internet access, power or water, and that the labs plan to brief Congress now to shape the crackdown they expect afterward.
Tech
Anthropic Splits Cyber Access Into Three Tiers, With Grid and Interbank Testing Reviewed Alongside the US Government
The expanded Cyber Verification Program gives vetted defenders reduced blocking classifiers on Claude Opus 5.5, Sonnet 5.5 and Mythos 5.1, and folds in Project Glasswing, whose partners logged 129,000 verified vulnerabilities from April to July.
Tech
Scale's Humanity's Sixth Sense Benchmark: Best Model Scores 53.6% on Intuitive Visual Reasoning Against 93.1% for Humans
Of 8,573 model failures, 94% came from perception or inference errors and only 5% from faulty logic, which points to the visual encoder, not the reasoning stack, as the bottleneck.
Tech
Microsoft Ships Decision-1, a Qwen-Based Scorer That Returns Calibrated Option Probabilities at $0.042 per Million Tokens
Microsoft claims 83.5% accuracy across 36 held-out benchmarks and median latency 35 times lower than GPT-6 Sol, but critics say the test sets are unnamed and one raw timing comparison shrinks to 1.4 times.
Tech
Berkeley Lab Puts Meta's SAM 3 and DINOv3 at the Center of a Genesis Mission Imaging Pipeline, Cutting Analysis From a Month to 15 Minutes
The multi-lab SYNAPS-I project runs fine-tuned versions of the open models on 300 A100 GPUs at national laboratory sites, starting with drought tracking in grapevine micro-CT scans.
Tech
Independent Researcher Uses Claude Code to Flag a 1.4-Earth-Radius Planet Candidate in Years-Old TESS Data, and NASA Approves Follow-Up
The candidate around TIC 4206066 repeats every 3.18 days and survived 74 agent-run falsification tests, but it is not yet a confirmed planet.
Tech
Anthropic Pauses Free Claude Team and API Credits for Startups After 21 Times the Usual Applications in 48 Hours
Hundreds of thousands of applications arrived within days of the October 6 launch, Anthropic admits that many non-startups were accepted, and it will re-verify everyone who has not yet claimed the benefits.
Macro
Anthropic's 2030 Economic Model Ranges From a 1.6% GDP Gain to a 32.4% Gain With Capital Taking 54.8% of Income
In the extreme scenario, knowledge-worker wages fall 11.5% below the no-AI path and nearly one in five knowledge workers is unemployed, while a survey of 10,980 Americans expects about a 10% GDP gain.
Tech
Trump Gives Musk, Huang and Brin the National Medal of Science as the White House Closes Ranks With US AI Champions
Trump called Elon Musk a "national treasure" and a modern-day Thomas Edison, completing Musk's return to the president's circle after their public split over tax and spending legislation.
Tech
Pitt's ARPA-H Robotic Wheelchair Runs Meta's DINO and SAM On-Device for Real-Time 360-Degree Perception
The up to $41.5 million RAMMP project pairs a robotic arm with open vision models optimized to run on battery power, without a cloud connection.