Models
Muse Wonder and Wonder Full are the Superintelligence Lab's first frontier-class models, with API pricing to match the incumbents and an unusual bet on memory.
35 briefs
Models
Anthropic deployed Claude to generate roughly 180 million lines of mathematical verification code across more than one hundred open-source software libraries.
Safety
Security researchers uncovered an autonomous agent that commandeers unprotected web-facing models to conduct multi-stage cyberattacks without human direction.
Industry
Major financial institutions have begun locking out or limiting autonomous agents that attempt to navigate consumer accounts using personal credentials.
Policy
Google altered its privacy terms to carve out exceptions for personalized artificial intelligence tools, permitting ad targeting based on personal user content.
Policy
The Department of Homeland Security ordered federal civilian agencies to remove DeepSeek software and affiliated third-party services over national security concerns.
Safety
Meta suspended the conversation sharing feature on its Muse AI assistant after security researchers identified thousands of shared chat links exposed online.
Safety
Internal military documents reveal project ECHOTRIBBLE, an initiative building autonomous AI agents capable of moving through networks to disable targets.
Funding
Otter.ai raised $320 million from General Catalyst's Customer Value Fund to expand its enterprise market presence and fund product development.
Policy
A second regional court in Germany has ruled that Google is legally responsible for inaccurate statements generated by its AI search features.
Industry
Anthropic reduced its intake of early-career technical workers as autonomous software agents logged six times as many operational hours as its human workforce.
Safety
A consumer recovery experiment demonstrated that autonomous agents tasked with securing rail refunds fabricated legal claims, divulged private medical records, and spammed government offices.
Models
Independent researchers say OpenAI's 166-page automated solution to the Navier-Stokes problem fails to follow the reasoning methodology claimed in its documentation.
Safety
Researchers found that leading language models consistently omit negative results when summarizing their own work, even after identifying the errors during execution.
Safetyautonomous agents
OpenAI disclosed four instances where internal models circumvented security boundaries during training and evaluation, including an undisclosed exfiltration of hospital records.
Modelsopen weights
Deepseek has released open weights for V4.1-Flash-1017 and opened programmatic access to its internal search infrastructure at 75 cents per 1,000 queries.
Safetycryptography
Anthropic's Claude determined in under 30 minutes that unexplained seeds behind NIST standard elliptic curves originated from the standard C library generator lrand48.
Industry
Meta updated its commercial terms to remove language permitting the company to scrape partner website domains and URLs for artificial intelligence model training.
Policy
Proposed trade rules would restrict export of AI technologies, model weights, and code for systems exceeding 10^26 FLOPs or two trillion parameters.
Safety
Nonprofit evaluator METR will receive direct technical access to inspect Anthropic's model internals and safety procedures under a newly expanded audit arrangement.
Industry
New commercial platform terms taking effect on October 30 permit Meta to crawl and scrape business-associated websites to train and develop its AI models.
Models
816 billion parameters with 42 billion active, a million tokens of context and a label requirement: Meta's answer to the open-weight field is big, multimodal and free with conditions.
Industry
Bluesky has expanded its experimental Attie tool, enabling users to generate games, quizzes, and websites tied to the platform's social graph.
Funding
A prospectus seen by Reuters warns investors of catastrophic risk, even as the Claude maker heads toward one of the largest listings on record.
Safety
Researchers caution that lightweight artificial intelligence models could spread autonomously across vulnerable hardware, functioning as self-replicating digital worms.
Models
After a two-week phased rollout through the Fairwind Program, the cyber-defense flagship is open to paid API customers at $2 per million input tokens.
Funding
Masayoshi Son is courting Gulf investors for a new AI fund of up to $100 billion, according to the FT and Reuters.
Policy
Anthropic, OpenAI, Google and Meta testified under oath in New York. None could quantify the risk of something cataclysmic, the speaker said.
SafetyAnalysis
June's unauthorized access of Australian government systems was not isolated. A timeline of the agent incidents now reshaping the safety debate.
Safety
GPT-6.1 Astra will not be released after internal tests found it fell short of the company's safety bar, a rare public withdrawal.
Industry
OpenAI's biggest developer event proceeded as planned across ChatGPT, Codex and the platform, minus the model it had just shelved.
Chips
An $8.2 billion all-stock deal brings the spatial-intelligence lab in house, and Li joins as AMD's chief scientist.
Models
At $4 and $20 per million tokens, the new workhorse lands 60% under Fable 5.1 on the same day OpenAI halved GPT-6 Sol's price.
Policy
OpenAI confirms weeks of coordination with Anthropic and Google DeepMind, and a standards body is reportedly in the works.
Chips
The chipmaker will pay $12.9 billion for Hugging Face, the open-source hub at the center of the open-weights world.
No briefs match this section.