Flash News

TREE NEWS update: AI startup Sierra and Meta have jointly launched the Personal Agent Protocol, an open standard governing how personal AI agents interact with businesses. Founding partners include Genesys, Instinct, Rocket, Shopify, Stripe and Walmart. The protocol uses OAuth for authentication, letting users grant agents read or write access to their accounts while businesses choose which functions to expose. Version 0.1 of the specification is planned for release later in October.

TREE NEWS update: Senior US Department of Justice officials have instructed employees to replace the term ‘artificial intelligence’ with ‘superintelligence’, abbreviated SI, in official communications. The directive applies to the department’s official correspondence. No further details on the scope or rationale were provided.

TREE NEWS update: Alphabet’s self-driving unit Waymo said it has launched fully autonomous driving in Detroit, expanding its driverless operations to the US auto industry’s home city. The company did not disclose the size of the initial service area, fleet size or a start date for public riders. Detroit becomes the latest US market where Waymo runs vehicles without a human safety driver.

TREE NEWS update: Hark has launched Hark Pro, an “agentic” AI app that performs tasks on users’ behalf such as buying groceries or booking a vehicle. The app works alongside the company’s AI hardware products, which are set to roll out starting in 2027. Hark Pro runs on Hark’s own cloud computer, Handoff, to remember user information and handle queries, and offers tools including action buttons and personalized mini-apps. It is available in a free version and paid subscription tiers.

TREE NEWS update: Google released Nano Banana 2.1, an image generation and editing model built on Gemini 3.6 Flash. The company said image quality, prompt adherence, character consistency and text rendering improved significantly over the previous generation, with output up to 4K. In preference testing, its thinking mode scored above 1050, versus 935 for Nano Banana Pro. The model is live in the Gemini app and API.

TREE NEWS reports: An analysis by SemiAnalysis found that, measured by API-equivalent value, Anthropic’s Opus 5.5 subscription is worth more than five times OpenAI’s GPT-6.1 Sol at the same tier. The firm said OpenAI recently halved the quota on its $200 Pro plan. Subscriptions account for only about 10% of Anthropic’s revenue but consume over 40% of its inference compute.

TREE NEWS update: Kansas City Federal Reserve President Jeff Schmid said AI is currently one of the biggest drivers of inflation. The remarks, made without further detail, place the technology at the center of the Fed’s inflation assessment. Schmid did not elaborate on the mechanism or on the implications for monetary policy.

TREE NEWS update: Anthropic launched a public beta of Claude for Google Workspace across all paid plans, letting users call the model from a sidebar add-on in Docs, Sheets and Slides. In Docs it can rewrite text in place or submit edits as suggestion cards; in Sheets it writes formulas, builds pivot tables and charts, and runs Python; in Slides it creates slides matching existing layouts. The add-on offers “review before edit” and “accept all” modes.

TREE NEWS update: Anthropic expanded its Claude Startups program on October 7 with the launch of Claude Startup Stack, offering eligible startups discounts and credits worth up to $45,000 from partners including Linear, Lovable, ElevenLabs, Granola and Hex. Applicants can also receive up to $7,000 in Claude products and credits, including one free year of Claude Team and $1,000 in API credits upon approval. Startups founded within five years or funded in the past two years are eligible.

TREE NEWS update: Paris-based Mistral has launched Large 4, a new AI model nicknamed “Le Chonk” after a June internet joke. The model tops GPT-6 Astra on one finance test but trails Claude on others. No release date, pricing or parameter count was disclosed.

TREE NEWS update: Google has signed a 20-year nuclear power agreement with Constellation Energy, the largest nuclear operator in the United States. The deal pairs the technology company with the country’s biggest reactor fleet operator for two decades of power supply. No financial terms were disclosed.

TREE NEWS update: The New Mexico Attorney General’s Office said federal regulators must revise their environmental assessment of a proposed natural gas pipeline built to power an Oracle data center. Attorney General Raúl Torrez called the original assessment inadequate because it examined only the pipeline’s impact and not the data center itself, and said regulators also failed to account for air pollution and greenhouse gas emissions from gas that would be flared during construction.

TREE NEWS update: OpenAI has expanded its partnership with Atlassian, under which OpenAI frontier models will power AI agents across Atlassian’s platforms and Rovo. The agreement broadens Atlassian’s access to OpenAI’s latest frontier models, including GPT-6 Astra and the GPT-5.6 series.

TREE NEWS update: Google released Gemini Embedding 2, a multimodal embedding model, the company announced. The model extends Google’s Gemini embedding line to handle multiple data modalities within a single embedding system. No pricing, availability details or benchmark figures were disclosed in the announcement.

TREE NEWS reports: Federal Reserve Bank of San Francisco President Mary Daly said she is concerned that AI, tariffs and energy costs could keep inflation elevated for longer. The remarks point to the cost pressures the Fed official sees persisting rather than fading quickly. No specific figures, policy proposal or timeline were given.

TREE NEWS update: Emil Michael, a member of the White House AI working group and chief technology officer at the Pentagon, said he has not yet seen anything regarding the Anduril contract. He added that he hopes there will be more than five prime contractors. Michael also said people should know AI can work for them, and that past AI incidents could have been prevented by laboratories.

TREE NEWS update: JPMorgan Chase CEO Jamie Dimon said Anthropic PBC’s AI model Mythos has sharply increased global cybersecurity risk, claiming on Tuesday that the risk posed by AI “increased 10-fold after Mythos appeared.” Dimon said AI has introduced vulnerabilities that were previously unknown, adding that cybersecurity was already a major concern before the model’s release.

TREE NEWS update: US Senator Elizabeth Warren sent letters to Treasury Secretary Scott Bessent and Commerce Secretary Howard Lutnick raising concerns about the impact of artificial intelligence on Trump administration policy. The letters were also addressed to Lutnick on the same issue. No further details of their contents were disclosed.

TREE NEWS update: Binance co-founder Yi He said at the exchange’s Smart Product launch AMA that Binance’s strategy includes going All in on AI, aiming for financial and information equality. Three AI products were unveiled: Finance AI for beginners, AI Pro for advanced traders, and AgentOS for developers. Finance AI will open to all users within seven days; AI Pro generates trading strategies via natural-language dialogue, while AgentOS lets developers connect large models to Binance’s trading, wallet and liquidity infrastructure, with the platform promoting selected agents to all users.

TREE NEWS reports: Hyperscale Data reported holding $56.4 million in bitcoin, cash and restricted cash, including nearly 220 BTC worth roughly $19 million, as of its latest disclosure. The AI data center company, formerly a bitcoin miner, said it is strengthening its balance sheet to support its business and develop its Michigan AI data center.