Here are today's news for 19th of October about Meta and Anthropic. Let's dive into the details. On October 12, 2026, Meta introduced Muse Wonder and Muse Wonder Full, its new flagship models for personal and professional AI, agent work, coding and reasoning. Muse Wonder launched through Meta’s API and Muse Code at $4 per million input tokens and $20 per million output tokens; Wonder Full is restricted to Muse Code subscribers and has no public API. Meta says the models can handle extended, multi, step tasks, recover from errors and use “Active Memory” to carry useful context across conversations. In independent Artificial Analysis testing, Wonder Full scored 56 on the Intelligence Index, eight points above Muse Spark 1.3; its Coding Agents Index score was 68. The release puts Meta more directly against OpenAI and Anthropic, though results vary by benchmark and provider. Developers and businesses gain another option for software and knowledge, work agents, while Muse users are expected to see Wonder features later, initially for subscribers. On 15 October 2026, the Financial Stability Board (FSB) published a final report setting out 12 “sound practices” for responsible AI adoption by banks and other financial institutions. Four cover firm, wide governance; eight address risk across AI development and deployment, from assessing use cases and governing data to testing performance, cybersecurity and third, party dependencies. The guidance is a practical toolkit, not a binding international standard, and firms are expected to tailor it to their risk profiles and existing controls. The warning is especially pointed for autonomous agents: the FSB says they may act unpredictably or outside intended boundaries, leading to unauthorized or erroneous actions, data, security incidents or disruption across connected systems. Human oversight should reflect a system’s autonomy and the materiality of its decisions; firms should also consider secure deployment and auditing agents’ actions. At the IMF, World Bank annual meetings in Bangkok on October 16, 2026, Anthropic co, founder and Head of Public Benefit Jack Clark renewed his call for binding safeguards on advanced AI. “It’s remarkable to me that the toys I buy for my children are regulated more than the literal super intelligence I work on at my day job,” he said, arguing that frontier development resembled “Formula One at night” without brakes or guardrails. Clark said Anthropic tests its systems, publishes results and shares information, but safety should not be left entirely to voluntary action: “We need common sense standards” to build confidence and support economic growth. His warning comes amid reported security incidents involving models from Anthropic and OpenAI, and a recent Anthropic report that small AI models could be months from worm, like self, replication across internet, connected devices. On October 8, the UK Information Commissioner’s Office said ten foundation, model developers, Amazon, Anthropic, Apple, Cohere, DeepSeek, Google, Meta, Microsoft, OpenAI and Stability AI, had made or committed to make data, protection changes after roughly two years of scrutiny. The measures include clearer explanations of personal, data use, stronger ways for people to exercise their rights, and tougher assessments of safeguards. The changes affect people whose information may be used in training data, and the companies and services built on their models; they also raise the compliance bar across a market where access to vast datasets is a competitive advantage. The ICO says developers must establish a lawful basis, explain processing, enable individual rights and show safeguards materially reduce risk. Its report also addresses sensitive data, such as health information and political opinions, and whether models can themselves contain personal data. Some firms have updated notices or assessments; others still owe changes. On October 15, Anthropic announced that six researchers from nonprofit evaluator METR had begun working with “employee, like access” to its systems, a move reported October 19. The team has Anthropic laptops, office space and internal tools broadly comparable to those used by the company’s risk, assessment staff. METR will assess Anthropic’s models, safety practices, risks and incidents, then publish its findings without Anthropic’s approval. The company may request redactions only for narrowly defined material, including customer data, privileged legal information and model, architecture details; METR may publicly judge whether a redaction was justified. The arrangement gives an outside evaluator unusual access to the people and systems behind frontier models, addressing a core weakness of external audits: limited visibility into training and internal risk work. That's all for today's AI news briefing. Have a great day!