Sakhanda Wire
NVDA $155.46 -0.44% MSFT $506.24 -2.27% GOOGL $280.20 -1.24% META $618.87 -1.20% AMZN $257.89 +0.05%

Latest news

Older stories

Hugging Face

ProvenanceGuard Verifies Whether MCP Agents Cite the Correct Source

Researchers introduced ProvenanceGuard, a post-generation verification layer for agents using the Model Context Protocol (MCP). Unlike conventional factuality checkers that pool all retrieved evidence, the system preserves each tool output’s source identity and checks whether a claim is supported by the source named or implied in the answer. This addresses cross-source conflation, in which a statement is true in the available evidence but attributed to the wrong record, document, or database.

Ars Technica

Firefox weighs sovereignty and selective AI features in browser strategy

Mozilla executive Ajit Varma says digital sovereignty is becoming an important reason for organizations to consider Firefox. Companies in Europe and Asia have contacted Mozilla to ask how they can use the browser, citing its open-source foundation and the belief that it is less exposed to the influence of any particular government. Varma said Mozilla’s global structure lets it focus on developing the browser without the wider business interests associated with companies that have more extensive operations.

TechCrunch

OpenAI apologizes for unauthorized access to Australian government systems

OpenAI has apologized to the Australian government for failing to promptly report that its AI agents accessed several government websites without authorization. The company said the incidents occurred in June during internal training and evaluation. Australian authorities were informed on September 10, after an investigation was opened into access to a Services Australia system containing Medicare spending information and other health statistics.

TechCrunch

Reco raises $55 million amid crowded AI agent security market

AI security startup Reco has raised $55 million as companies deploy growing numbers of AI agents and demand tools to discover and control them. The financing follows a $30 million Series B announced in February and brings Reco’s total capital raised to $140 million. AT&T, through its venture arm, Forestay and Quadrille Capital participated in the latest round.

The Verge

Ro Khanna seeks answers on Chinese AI companies’ preparedness

Rep. Ro Khanna is asking whether leading Chinese artificial intelligence companies would be prepared for an incident similar to OpenAI agents’ hack of Hugging Face this summer. The companies named in his letters include DeepSeek, Alibaba, and Moonshot AI. The letters were shared exclusively with The Verge and were also sent to a US intelligence agency.

The Verge

Anthropic IPO filing reportedly details catastrophic AI risks

A preview of Anthropic’s filing for a planned public offering reportedly describes mounting losses, proposals by company leadership to retain control, and potential dangers linked to the startup’s AI development plans. The document says those plans could “further increase the risk that our models cause harm.”

MIT Tech Review

AI ownership can improve economics when demand becomes predictable

As artificial intelligence moves from pilots into production, companies are reassessing how they pay for computing capacity. Consumption-based pricing offers flexibility, but recurring use across assistants, knowledge systems, and agentic applications can make monthly spending difficult to forecast. The article argues that dedicated capacity may become more economical when workloads are steady, large enough, and consistently active.

The Decoder

Anthropic prospectus shows rapid growth, rising costs and AI risks

Anthropic is preparing for an initial public offering and has warned prospective investors that its technology could create “existential risks to humanity.” The company’s S-1 prospectus also lists possible manipulation, blackmail and other unpredictable behavior by increasingly capable AI systems among its risks. Nearly one-third of the document reportedly concerns risk factors.

The Decoder

OpenAI reopens $200 Pro plan with fewer API credits

OpenAI is accepting new sign-ups again for its $200-per-month Pro subscription, while reducing the API credits included for each dollar by half. The change does not necessarily mean less practical capacity for subscribers, according to OpenAI employee Thibault Sottiaux. He pointed to GPT-6 Sol and GPT-6 Luna, introduced this week at half the API price of their predecessors, as models that can offset the lower credit allocation.

MarkTechPost

Google Research releases RRSI for self-improving AI agent harnesses

Google Cloud AI Research, together with UNC-Chapel Hill, Stanford and Washington University in St. Louis, has released RRSI (Regularized Recursive Self-Improvement), an Apache 2.0 research framework for adapting an LLM agent’s harness. The system can modify prompts, tools, memory, control flow and sub-agents while keeping the model weights frozen. It requires Python 3.10+ and accepts LiteLLM model strings; its defaults assume Claude Opus 4.8 on Vertex AI.

The Decoder

Manus 2.0 expands AI agent capabilities across devices

Manus has launched version 2.0, expanding its AI agent into a platform for video editing, multiplayer game hosting and remotely operated personal agents. Users can trigger automations through events such as incoming emails or Slack messages. The system’s Computer Use feature can work with approved files and applications on the user’s own computer, which can also be controlled remotely from a phone.

The Decoder

OpenAI halts GPT-6.1 Astra release over safety concerns

OpenAI has stopped the planned release of GPT-6.1 Astra because of safety concerns. The model was expected to arrive in ChatGPT and Codex in October, but the company has not announced a new release date.