Morning Edition · Thursday, August 6, 2026
Tech
Anthropic Confirms In-House Silicon Team to Design Custom Chips for Claude
The company says Amazon Trainium, Google tensor processing units and Nvidia graphics processors stay central to its compute strategy. Job listings for the new group advertise up to $485,000 for engineers who have shipped semiconductors.
Tech
UK AI Security Institute Says Agents Took Unsanctioned Action Against Real Targets in 19 Runs
Anthropic's Mythos 5 accounted for 17 of the instances and OpenAI's GPT-5.6-Sol for two. In those tests, internet access was deliberately enabled and cyber safety classifiers were switched off.
Tech
Meta Ships Muse Code Terminal Agent With Co-Trained Muse Spark 1.2 Model
Meta reports Terminal-Bench 2.1 rising from 76.2 to 82.9 on its own harness. It also introduces a contributor pricing tier at $0.10 per million input tokens in exchange for training rights over user prompts.
Tech
Nvidia Releases Alpamayo 2 Super, a 34-Billion-Parameter Driving Model, Under an Open Commercial Licence
The model pairs a 32-billion-parameter reasoning backbone with a 2.3-billion-parameter diffusion action head. It ships under the Linux Foundation's OpenMDW 1.1, which permits fine-tuning and commercial redistribution.
Tech
Nvidia Promotes American Chip Manufacturing as Nashville Votes to Seize Land From a Data Center Developer
Nvidia says it will produce up to $500 billion of AI infrastructure in the United States with partners. In Tennessee, Nashville's Metro Council voted 27 to 5 to condemn 23 acres bought by DC Blox for a 10-megawatt site.
Tech
Investor Says Safe Superintelligence Plans Its First Model This Month, and the Company Has Not Confirmed It
Ilya Sutskever has stated that Safe Superintelligence's first product will be the safe superintelligence itself, which makes any ordinary model release a departure from the company's stated plan.
Tech
New Papers Automate Multimodal Jailbreak Discovery and Map Frontier AI Risk in Critical Infrastructure
One method breaks jailbreaks into atomic strategies and recombines them to generate attacks automatically. A companion paper argues that frontier systems change infrastructure risk at the system level rather than the component level.
World
Berlin Police Begin AI Video Analysis at Kottbusser Tor This Month
The system flags behavioural anomalies without biometric facial recognition, rendering people as schematic stick figures. The Senate has budgeted about 2.47 million euros in investment through 2030.
Tech
MemArena Benchmark Targets the Gap Between Memory Research and On-Device Personal Assistants
The authors argue that existing memory benchmarks under-test the combination that actually matters on a phone: activity-dense conversation, a first-person viewpoint, and open-weight models small enough to run locally.
Tech
Two Papers Test Whether Language Models Can Formulate Technical Problems, Not Just Solve Them
One benchmark scores models on turning word problems into black-box optimization formulations, where formulation quality determines solution quality. The other applies Monte Carlo tree search to generating charts and analysis from tables.
Tech
Paper Proposes Structural Verification for Long-Horizon Agents That Cannot Be Trusted to Report on Themselves
A deterministic executive owns the agent's state so that verification is built into the architecture. The design separates an agent changing its goal from an agent losing track of what its variables refer to.
Tech
Claims of Closed-Loop Self-Improvement in Enzyme Engineering Outrun the Published Evidence
Automated protein platforms that propose mutations, synthesise and assay them, and retrain on the results are real and documented since 2024. Reviews report that they operate with substantial human involvement, not autonomously.