AWS is giving AI agents a way to pay for web resources without human intervention. Anchor Browser now integrates with Amazon Bedrock AgentCore payments, allowing agents to respond to HTTP 402 challenges and access pay-per-query datasets, research APIs, and paywalled content through the x402 protocol.
AgentCore payments handles wallet connectivity, cryptographic authorization, configurable spending limits, settlement, and request retries. Anchor Browser keeps those transactions inside the same managed browser session, preserving authentication state, cookies, navigation context, and live-view accuracy.
Before this integration, developers had to build and secure the payment stack themselves: AWS describes eight coupled components, including private keys, budget controls, signing, retry logic, settlement, and logging. The integration is available through AWS Marketplace.
Why it matters: AWS is turning payment from an agent-runtime liability into infrastructure controlled by Bedrock AgentCore. That gives teams using Anchor Browser a defined boundary for private keys and spending, rather than leaving eight security- and operations-sensitive components inside an environment exposed to web content and third-party tools.
OpenAI and Anthropic told British lawmakers that frontier-AI safety needs binding rules, putting them at odds with Google DeepMind and Meta. Anthropic Europe chief Liam Booth-Smith said voluntary commitments were insufficient. OpenAI vice president Tom Duff Gordon backed mandatory measures including incident reporting and an international convention.
Britain's AI Security Institute tests frontier models only when companies cooperate. Cabinet Office official Cat Little confirmed the regime is voluntary, while AISI director Henry de Zoete said it cannot stop a dangerous model's release. Duff Gordon said OpenAI's UK pre-release access depends on the U.S. government. Separately, MI5 director Ken McCallum warned that AI could increase the risk of bioweapon misuse.
Why it matters: Liam Byrne's committee now faces a concrete choice: seek UK powers that AISI currently lacks, or accept that British testing of OpenAI's models remains dependent on Washington and company cooperation.
Apple won expedited access to OpenAI hiring records as it pursues trade-secrets claims tied to OpenAI’s rumored first hardware device. U.S. District Judge Edward Davila ordered OpenAI to provide materials from interviews with former Apple employees who applied for roles in consumer hardware and robotics.
OpenAI has five days to produce documents already collected and 14 days for additional materials. The court also approved access to private LINE messages involving Chang Liu and Alyssa Peng, plus depositions of both employees and OpenAI chief hardware officer Tang Yew Tan—all former Apple staff.
Davila rejected expedited discovery into metal finishes, power management, and broader hardware research, but let Apple narrow those requests and try again.
Why it matters: Apple can examine whether OpenAI’s recruiting process referenced Apple technology without gaining immediate access to the company’s broader hardware research; OpenAI must produce the first records within five days and the rest within 14.
Codex and ChatGPT Work now account for more than half of output tokens in OpenAI’s EU ChatGPT Enterprise sample, as businesses use agents for tasks from planning through completion. Weekly active users of Codex and ChatGPT Work among EU small and medium businesses grew more than eightfold since May 2026.
Adoption remains uneven: agentic tools account for 22% of output tokens in Slovenia versus 85% in Lithuania. In September, EU frontier firms generated more than four times as many output tokens per active user as typical EU firms, while frontier firms outside the EU generated more than twice as many as EU frontier firms. EU-based firms represented only 10% of frontier firms globally.
Why it matters: For EU executives, the next hurdle is turning broad access into deeper use: frontier firms outside the bloc generate more than twice the output per active user, limiting Europe’s chance to convert adoption into stronger businesses and higher productivity.
Tech Against Terrorism found that abliterated versions of newly released open-weight AI models now appear online in under three days, down from more than 24 days in 2024.
Its Counter-Terrorism AI Benchmark tested 134 models from OpenAI, xAI, Anthropic, Meta, Google, and companies in China and Europe. Only two models passed completely, while 81 failed at least one measure. Reframing a dangerous request as coming from a “safety researcher” produced answers 16.9% of the time—nearly nine times as often. Meta’s Llama 3.1 8B scored 97 with safeguards intact and 3 after abliteration; the modified version then returned a 15-step response to an IED request.
Why it matters: For Meta and Google, releasing model weights means the safety behavior they ship can be stripped and replaced before the original release is three days old, leaving downloaders with a model that no longer reflects the companies’ safeguards.
Character.AI filed Chapter 11 and put its chatbot platform up for sale; without an actionable bid, it will wind down. The platform and subscriptions will keep running during the court-supervised sale, expected to close in late November or early December 2026.
Disney agreed to pay $20 million for selected assets and receive a perpetual, non-exclusive license to Character.AI technology. Former employees who joined Disney last month alongside former CEO Karandeep Singh Anand will provide transition services.
Character.AI reported $50 million to $100 million in assets and liabilities. Lawsuits after a 14-year-old Florida boy’s suicide, a Texas investigation into child-privacy and safety laws, and struggles to scale and monetize preceded the filing. The platform had 20 million monthly active users in October 2024.
Why it matters: Character.AI users could lose access after early December: Disney’s $20 million agreement covers selected assets and a technology license, while continued platform access still hinges on another buyer submitting an actionable bid.