Here are today's news for 29th of October about IBM, Meta, Anthropic and Google. Let's dive into the details. On October 27, 2026, at TechXchange in Atlanta, IBM announced identity, access and monitoring controls for enterprise AI agents. IBM Agent Identity, now generally available, records each agent’s owner, lifecycle status and authorized scope; paired with HashiCorp Vault 2.2, it supports least, privilege credentials and real, time revocation when an agent exceeds permission. IBM says watsonx Orchestrate can discover agents, models and MCP servers across Amazon Bedrock, Microsoft Foundry and Google Gemini, run agents in isolated sandboxes, and evaluate them before deployment and monitor them in production. Cost tracking exposes model and token use; Trace Inspector shows execution steps and tool interactions. The launch addresses agent sprawl: security and IT teams need to know which automated actors can reach business systems and be able to trace or stop their actions. IBM is positioning Orchestrate as a cross, vendor management layer rather than a platform limited to its own agents. On Oct. 26, Meta announced plans to raise as much as $95 billion for AI infrastructure and development: an initial $35 billion in offerings, about $22.5 billion of Class A common stock and $12.5 billion of depositary shares for mandatory convertible preferred stock, plus a proposed $60 billion at, the, market sale of new, nonvoting Class C shares. The larger sale is expected no earlier than Q1 2027; its terms are not final, and the full amount is not guaranteed. Nvidia, AMD and Broadcom indicated interest in buying up to $5 billion of the common shares combined, but their interest is nonbinding. The scale signals that building AI compute is becoming a major capital, raising contest, not just a software race. Meta says the proceeds will support its apps, AI agent Muse and “personal superintelligence”; extra capacity could bolster products and advertising while intensifying competition with OpenAI and Anthropic. Meta expects 2026 capital expenditure of $135 billion to $145 billion, amid its push to expand data centers and computing capacity. Anthropic’s Oct. 26 IPO filing put hard numbers behind Claude’s surge: revenue reached $16.3 billion in the six months to June 30, up from $1.16 billion a year earlier, while net loss hit $80.7 billion. The headline loss is misleading: $72.4 billion reflected changes in the fair value of convertible notes and related instruments; operating loss was $693 million, and adjusted operating income was $1.62 billion. In Q2, Anthropic reported $11.6 billion revenue and $5.6 billion in compute costs, $2.08 of revenue per compute dollar. The filing also warned that advanced systems could act autonomously, evade oversight or display “self, preserving behaviors,” and that AI, assisted research may outpace safety measures. It groups AI threats as misuse, loss of human control, authoritarian abuse and economic disruption. Those disclosures matter to investors and customers as Anthropic competes with OpenAI and Google while depending on costly cloud and chip infrastructure. On October 27, 2026, Google DeepMind announced Genie 4, a real, time world model now rolling out as a research preview to trusted testers. It generates interactive, photorealistic scenes with native audio: ambient sound, music and dialogue that responds to users and events, rather than sound added after video generation. Users can steer a running simulation with text, changing elements such as weather or camera movement; Google says characters move and react more naturally than in Genie 3, unveiled in August 2025. The advance could help game developers and other creators prototype responsive virtual environments, while raising the bar for competitors whose video tools primarily produce fixed clips. DeepMind says Gemini 3.8 Flash helps guide simulations by interpreting scenes and breaking goals into steps. The company has not announced Genie 4 pricing or public, release dates; planned access includes developers and Google AI Ultra subscribers, whose plans start at $100 monthly. On October 26, 2026, OpenAI announced three research results from internal models: a proof of the local smoothing conjecture in harmonic analysis, a claimed refutation of computer science’s Exponential Time Hypothesis (ETH), and an existence proof for certain four, dimensional quantum field theories, including quantum chromodynamics. The claims are not yet independently verified. OpenAI published the math output and Lean formalization; for the ETH result, it released a 67, page manuscript and Lean proof, while an external group led by Russell Impagliazzo and Ramamohan Paturi of UC San Diego and Ryan Williams of MIT will examine it. For physics, OpenAI is leaving release decisions to a group led by Nobel laureate David Gross. The ETH claim is especially consequential: if correct, it overturns a widely used assumption about the limits of efficient algorithms. OpenAI called its manuscripts “raw research data” and said it does not yet know which review model works best. That's all for today's AI news briefing. Have a great day!