Latest AI News

Airbnb CEO Brian Chesky announced plans to build AI-native features into the platform, including LLM-powered search, trip planning, and host management tools. A natural language search feature is live to a small percentage of users, with plans to make it more conversational. The existing AI customer support bot now resolves a third of issues without human help, with voice support and expanded language coverage planned. New CTO Ahmad Al-Dahle, formerly of Meta's Llama team, will lead AI efforts.

Disney sent a cease-and-desist letter to ByteDance over its new Seedance 2.0 AI video model, accusing the TikTok parent of stocking the platform with pirated Disney IP spanning Star Wars, Marvel, and Family Guy. The letter, first reported by Axios, calls ByteDance's actions a "willful, pervasive" smash-and-grab. The MPA and Human Artistry Campaign, whose members include SAG-AFTRA and the DGA, also condemned Seedance 2.0 after deepfakes of Tom Cruise and Brad Pitt went viral.

The Pentagon's War Department has designated Anthropic a national security supply-chain risk, banning all defense contractors from using Claude AI. The trigger was an Anthropic executive contacting Palantir after a U.S. raid to capture Venezuelan President Nicolas Maduro, asking whether Claude was used — effectively probing for classified information. Palantir, which used Claude as the sole model in certain classified Maven Smart Systems workflows, must now rebuild those integrations. Anthropic CEO Dario Amodei has vowed to challenge the designation in court.

Nvidia researchers have developed Dynamic Memory Sparsification (DMS), a technique that compresses the KV cache in large language models by up to 8x, cutting reasoning costs without accuracy loss. Unlike heuristic methods that discard tokens using fixed rules, DMS trains models like Llama 3 or Qwen 3 to intelligently decide which tokens to keep or evict, using a "delayed eviction" mechanism. On AIME 24 math benchmarks, a Qwen-R1 32B model scored 12 points higher than standard models under equal memory budgets. DMS is available in Nvidia's Model Optimizer framework.

Microsoft AI CEO Mustafa Suleyman told the Financial Times that AI will reach human-level performance on most white-collar tasks within 12 to 18 months, predicting roles like lawyers, accountants, and project managers will be fully automated. He noted AI-assisted coding is already transforming software engineering. Anthropic CEO Dario Amodei has similarly warned AI could eliminate half of entry-level white-collar jobs. Sen. Bernie Sanders called Suleyman's prediction an economic earthquake and renewed calls for a moratorium on AI data centers.

ElevenLabs has become the first company to secure an AIUC-backed insurance policy covering AI voice agents. The certification subjects AI systems to over 5,000 adversarial simulations spanning data privacy, safety, security, and reliability. ElevenLabs' ElevenAgents platform powers over three million voice agents used by more than 75% of Fortune 500 companies, including Cisco and Revolut. The policy lets enterprises insure against AI agent errors, such as incorrect customer information. Over 95% of enterprise AI pilots fail to reach deployment due to legal and security concerns.

Tavus has launched Raven-1, a multimodal AI system designed to interpret human emotion and intent in real time. The system is part of Tavus's broader push toward emotionally intelligent AI, alongside the newly live Phoenix-4, described as a real-time human rendering model with emotional intelligence. Tavus offers developer API access for teams building human-like AI experiences, as well as a PALs platform for personal AI companions. The launches signal Tavus's focus on making AI interactions more emotionally aware and contextually responsive.