Morning Edition · Saturday, July 11, 2026
Tech
Meta Puts Its Frontier Model Behind a Paid API, Closing the Weights on Muse Spark 1.1
The multimodal agentic model comes with a one-million-token context window and metered pricing of $1.25 per million input tokens, ending Meta's open-weight approach at the frontier.
Tech
Meta Moves Iris Accelerator Into Production as It Targets 14 Gigawatts of Compute
The in-house chip, designed with Broadcom and fabricated by TSMC, cleared testing in six weeks and enters production in September amid roughly $145 billion in projected annual infrastructure spending.
Tech
Anthropic and AE Studio Build a Removable Compartment for Dangerous Knowledge
The method, called Gradient-Routed Auxiliary Modules, removed virology and cybersecurity capabilities from a test model about as completely as never training on the data at all, without reducing general performance.
Geopolitics
Anthropic Redeploys Fable 5 After a 19-Day Export Suspension
The model returns with a classifier, a filter that detects the attack, which Anthropic says blocks the triggering jailbreak more than 99 percent of the time, alongside an industrywide framework for scoring jailbreak severity.
Tech
Meta's Non-Invasive Brain-to-Text System Decodes Typed Sentences From Magnetic Signals
Brain2Qwerty reaches a 32 percent character error rate using magnetoencephalography (MEG) and 19 percent for the best participants, with no surgery, though electroencephalography (EEG) remains far worse at 67 percent.
Tech
Deutsche Telekom Puts ChatGPT Enterprise in Front of 200,000 Employees
The German carrier is adding live translation, in-call assistants, and post-call summaries to its existing voice and messaging channels, with dedicated computing hosted inside Germany to keep data in the country as the law requires.
Tech
Meta Tests Always-On "Super Sensing" Glasses That Recall a Wearer's Day
Prototype glasses would continuously sample audio and images and upload the extracted metadata for an assistant to search, reviving a dispute over recording indicators and consent.
Tech
ZipDepth Distills a Vision Foundation Model Into a 6-Million-Parameter Depth Network
The compact model converts into plain convolutions with no custom operators, keeping its zero-shot accuracy while running in real time on GPUs, CPUs, and embedded hardware.
Tech
A Research Proposal Would Make Enterprise Agents Act Before Being Asked
A new paper argues that retrieval and agent frameworks keep AI agents fundamentally reactive, and it proposes context graphs to let them anticipate and start work on their own.
Tech
Anthropic Researchers Repeat the Claim That Today's Models Can Automate White-Collar Work
The claim that current systems are enough to automate white-collar jobs within five years, even if progress stalls, conflicts with the same group's data showing that actual usage is far below what the models can theoretically do.
Tech
Anthropic Details How Claude Code Went From Internal Tool to Product
The account traces the coding agent's path from an internal command-line tool to a shipped product, published as rival developers compete to match Anthropic on code.
Tech
Russian Lab Proposes a Unified Way to Compare LLM Fine-Tuning Methods
Presented at the International Conference on Machine Learning (ICML) 2026, the work sets out a common framework for evaluating offline preference-based fine-tuning methods, which learn from prepared answer pairs.