Latest AI News

OpenAI is expanding its Trusted Access for Cyber program to thousands of verified individual defenders and hundreds of teams protecting critical software. The company is also introducing GPT-5.Cyber, a fine-tuned variant of GPT-5.4 with fewer capability restrictions and lower refusal boundaries for legitimate cybersecurity work, including binary reverse engineering to analyze compiled software for malware and vulnerabilities. Initially limited to vetted security vendors and researchers, individuals can apply at chatgpt.com/cyber while enterprises can request access through OpenAI representatives.

Boston Dynamics demonstrated its Spot robot performing household cleanup tasks — picking up shoes and soda cans — using Google's Gemini Robotics-ER 1.5 visual-language model. The project, which grew from a 2025 internal hackathon, used natural language prompts instead of traditional state-machine programming. Engineers built an SDK layer connecting Gemini Robotics to Spot's API, limiting actions to navigation, image capture, object identification, grasping, and placement. Boston Dynamics has since rolled out AIVI-Learning powered by the newer Gemini Robotics ER 1.6 for Spot and Orbit.

NVIDIA has launched Ising, the world's first family of open source AI models designed to accelerate quantum computing by tackling calibration and error correction. Ising Calibration uses a vision language model to reduce processor calibration time from days to hours, while Ising Decoding delivers up to 2.5x faster and 3x more accurate error correction than pyMatching, the current open source standard. Adopters include Fermilab, Harvard, IQM Quantum Computers, and Lawrence Berkeley National Laboratory.

AWS has launched Amazon Bio Discovery, an agentic application designed to make lab-in-the-loop drug discovery accessible to all researchers. The platform provides access to 40+ AI biology models with AI-guided selection, a no-code workflow environment, and integration with contract research organization partners for wet-lab validation. In a collaboration with Memorial Sloan Kettering Cancer Center, the tool was used to design nearly 300,000 novel antibody candidates and filter them to the top 100,000 for wet-lab testing in weeks rather than up to a year.

Anthropic tested nine instances of Claude Opus 4.6 as Automated Alignment Researchers (AARs) on the weak-to-strong supervision problem, a proxy for overseeing smarter-than-human AI. Human researchers achieved a performance gap recovered score of 0.23 over seven days. The AARs reached a PGR of 0.97 in five additional days, spending roughly $18,000 in compute. Methods generalized to math tasks with a 0.94 PGR but only 0.47 on coding. The approach failed to show significant gains at production scale with Claude Sonnet 4.

Anthropic has redesigned the Claude Code desktop app to support parallel agentic workflows. The update introduces a new sidebar for managing multiple sessions across repos, filterable by status, project, or environment. New features include an integrated terminal, in-app file editor, faster diff viewer, drag-and-drop pane layout, and expanded HTML and PDF preview. SSH support now extends to Mac. The redesign is available to Pro, Max, Team, and Enterprise plan users and via the Claude API.

Anthropic has launched Claude Code Routines in research preview, letting developers configure repeatable automations that run on a schedule, via API call, or in response to GitHub repository events. Each routine bundles a prompt, repo, and connectors, and runs on Claude Code's cloud infrastructure without requiring a local machine. Use cases include nightly bug triage, deploy verification, and automatic PR reviews. Available to Pro, Max, Team, and Enterprise users, with daily limits of 5, 15, and 25 routines respectively.

Microsoft has launched MAI-Image-Efficient, a new text-to-image model available now in Microsoft Foundry and MAI Playground. The model delivers flagship-quality image generation at 41% lower cost than its predecessor, priced at $5 per 1M text input tokens and $19.50 per 1M image output tokens. It is 22% faster and 4x more efficient than MAI-Image-2, and 40% faster on average than competing models. Shutterstock is already testing it, and rollout to Copilot, Bing, and PowerPoint is planned.

Character.ai has launched PipSqueak 2 (PSQ2), a major model upgrade replacing the current PipSqueak, available now for c.ai+ subscribers and rolling out to free users in early May. PSQ2 improves in-character consistency, context retention, and writing quality. The platform also upgraded its Memory system with expanded tracking categories, in-chat notifications when memories are recorded, and a coming Memory Visualization meter. Additionally, Character.ai previewed Lorebook, its most-requested feature, which lets creators attach world-building lore entries to characters for richer, more consistent storytelling.

Google DeepMind has launched Gemini Robotics-ER 1.6, an upgraded embodied reasoning model for robotics that improves spatial reasoning, object counting, multi-view success detection, and tool use including native Google Search calls. A new instrument reading capability, developed with Boston Dynamics, lets robots like Spot interpret pressure gauges and sight glasses with 93% accuracy using agentic vision. The model outperforms both Gemini Robotics-ER 1.5 and Gemini 3.0 Flash and is available now via the Gemini API and Google AI Studio.

Google has launched its desktop app for Windows globally in English, featuring built-in AI Mode search. Announced by Group Product Manager Vinay Mahagaokar, the app lets users access Google Search, web results, local files, installed apps, and Google Drive via a keyboard shortcut (Alt + Space). It also includes screen sharing for contextual questions and Google Lens for searching anything on-screen, such as translating images or getting homework help.

Google has added a new prompt autocomplete feature called Tab Tab Tab to AI Studio's vibe coding experience. Powered by Gemini, the tool helps users flesh out vague or incomplete ideas by automatically filling in the blanks when writing prompts. The feature is designed to lower the barrier for users who arrive with rough concepts rather than fully formed instructions, making AI-assisted coding more accessible to beginners and casual developers.

OpenAI chief revenue officer Denise Dresser sent a four-page internal memo urging employees to build an enterprise moat and pursue multi-product adoption to reduce customer churn. Dresser, who recently took over duties from former COO Brad Lightcap, wrote that the market is as competitive as she has ever seen it and called out rival Anthropic directly, accusing it of inflating its run rate and making a strategic misstep by not acquiring enough compute. The memo frames OpenAI as a platform company with multiple entry points.

Meta is training an AI avatar of CEO Mark Zuckerberg to stand in for him at employee meetings, according to a Financial Times report. The avatar is being trained on Zuckerberg's image, voice, mannerisms, tone, and public statements so employees feel more connected to the founder. Zuckerberg is personally involved in the training and now spends five to ten hours per week coding on Meta AI projects. If successful, Meta may expand the feature to let creators build AI avatars of themselves.

Microsoft is testing OpenClaw-style always-on AI agents for its 365 Copilot platform, according to The Information. Corporate VP Omar Shahine confirmed the company is exploring OpenClaw technologies in an enterprise context. The autonomous agents would run around the clock, monitoring Outlook inboxes and calendars and suggesting daily tasks. Microsoft is also developing role-specific agents for marketing, sales, and accounting to limit permissions. Microsoft plans to reveal features at its Build conference starting June 2nd, and says it can implement safer versions of the tool.

Cloudflare has integrated OpenAI's GPT-4.5 and Codex models into its Agent Cloud platform, giving millions of enterprise customers access to frontier AI for agentic workflows. Built on Cloudflare Workers AI, Agent Cloud lets businesses deploy agents that handle tasks like customer responses, system updates, and report generation in a secure, production-ready environment. The Codex harness is now generally available in Cloudflare Sandboxes, with Workers AI support coming soon. OpenAI APIs currently process over 15 billion tokens per minute.

Danish drugmaker Novo Nordisk has announced a partnership with OpenAI to apply AI across drug discovery, manufacturing, supply chains, and corporate operations. The maker of Wegovy and Ozempic said pilot programs will begin across research and development, manufacturing, and commercial operations, with full integration planned by end of 2026. OpenAI will also help train Novo's global workforce in AI literacy. Novo CEO Mike Doustdar and OpenAI CEO Sam Altman both cited the potential to accelerate treatment discovery for obesity and diabetes. Financial terms were not disclosed.

Blackmagic Design announced DaVinci Resolve 21 at NAB 2026, introducing a new Photo page that brings Hollywood-grade node-based color tools to still photography. New AI features include IntelliSearch for content and face searching, CineFocus for depth-of-field adjustment, Face Age Transformer, Face Reshaper, Blemish Removal, AI UltraSharpen, and Slate ID for automatic metadata extraction. Fairlight gains folder tracks for audio management, and Krokodove adds over 70 new Fusion graphics. The public beta is available now from Blackmagic Design's website.

OpenAI CEO Sam Altman's San Francisco home was targeted in a second attack early Sunday morning, just two days after a Molotov cocktail incident. Police arrested Amanda Tom, 25, and Muhamad Tarik Hussein, 23, after surveillance footage showed a passenger in a Honda sedan firing a gun at the Russian Hill property around 1:40 a.m. Three firearms were found at their residence. The first attack on Friday involved Texas man Daniel Alejandro Moreno-Gama, who was booked on attempted murder and arson charges.

MiniMax has open-sourced its M2.7 model, which achieves state-of-the-art performance on two coding benchmarks: SWE-Pro with a score of 56.22% and Terminal Bench 2 with a score of 57.0%. The model is now available on Hugging Face, with API access offered through the MiniMax platform. The release positions MiniMax as a competitive player in the open-source coding AI space, where benchmark performance on software engineering tasks is a key differentiator.

OpenAI CEO Sam Altman shared a family photo on his blog after someone threw a Molotov cocktail at his home at 3:45 am, though it bounced off and no one was hurt. Altman linked the attack to a recent incendiary article about him, saying he had underestimated the power of words. He used the post to outline his AI beliefs, including democratizing access, avoiding power concentration, and prioritizing safety, while reflecting on mistakes during his decade leading OpenAI.

CoreWeave (Nasdaq: CRWV) has signed a multi-year agreement with Anthropic to support the development and deployment of Anthropic's Claude family of AI models. Compute will come online starting later this year, with an initial phased infrastructure rollout that has potential to expand over time. Anthropic will run production-scale workloads on CoreWeave's cloud platform. The deal makes nine of the top ten AI model providers CoreWeave customers, highlighting surging demand for large-scale AI infrastructure.

VP JD Vance and Treasury Secretary Scott Bessent held a private phone call with top tech CEOs — including Anthropic's Dario Amodei, Google's Sundar Pichai, OpenAI's Sam Altman, and Microsoft's Satya Nadella — to discuss AI model security ahead of Anthropic's Mythos release. The call focused on large language model security posture, safe deployment, and how to respond if models scale in favor of attackers. Anthropic later rolled out Mythos to limited partners including Apple, Google, and Microsoft.

Regal Cineworld has launched the first dedicated movie ticketing app inside ChatGPT, developed with The Boxoffice Company, making it the first major U.S. exhibitor to do so. Users can ask conversational prompts about nearby showtimes, formats, and screening options, then are directed to Regal's website to buy tickets. The tool, built on The Boxoffice Company's Boost platform, covers Regal's 394 U.S. locations and 5,386 screens, aiming to streamline the path from film discovery to purchase.

Pika Labs has launched a cash-back monetization system for its Pika AI Self agents, allowing creators to earn real money when others interact with their agents or use their skills. The system uses redeemable tokens that accumulate each time someone engages with a creator's agent. Pika frames the feature as making agents work for their creators in a literal financial sense, expanding the platform beyond content generation into creator monetization.

Anthropic has made Claude Cowork generally available on all paid plans, adding enterprise controls for organization-wide deployment. New features include role-based access controls with SCIM support, group spend limits, usage analytics via dashboard and Analytics API tracking DAU/WAU/MAU, and expanded OpenTelemetry support compatible with Splunk and Cribl pipelines. A new Zoom MCP connector brings meeting summaries and action items into Cowork workflows. Admins can also restrict per-tool connector permissions org-wide from the admin console.

Perplexity has expanded its Plaid integration to let users securely link bank accounts, credit cards, mortgages, auto loans, and student loans alongside previously supported brokerage accounts. The read-only connection covers 12,000+ financial institutions including Chase, Fidelity, and Schwab, with user data never touching Perplexity's servers. Using Perplexity Computer, users can analyze spending, calculate net worth, and build custom dashboards or debt payoff plans via freeform questions. The feature is live on desktop for US and Canada users, with advanced tools limited to Pro and Max subscribers.

Upwork has launched a ChatGPT app that lets businesses describe project needs and find and hire talent from the platform's 18 million professionals without leaving the chatbot. Users can draft job posts inside ChatGPT, then move to the Upwork Marketplace to handle compliance, payments, and contracts, with Upwork's AI agent Uma available for additional help. The integration supports OpenAI's push to make ChatGPT an all-in-one hub, which already serves over 900 million weekly users.

OpenAI is updating its ChatGPT subscription tiers to better accommodate growing demand for Codex, its AI-powered coding tool. The company is introducing a new Pro tier priced at $100 per month, which offers five times more Codex usage than the existing Plus plan. The Pro tier is designed for longer, high-effort Codex sessions, making it well-suited for power users who rely on extended coding workflows. The update signals OpenAI's push to serve more intensive developer use cases within ChatGPT.

A D.C. federal appeals court denied Anthropic's request to pause the Pentagon's supply-chain risk designation, dealing a setback to the AI company despite a separate San Francisco court granting a preliminary injunction last month blocking bans on Claude. The split rulings mean Anthropic still faces ongoing legal battles. The Pentagon can still exclude Anthropic from new contracts under a separate statute, though it is expected to continue using Anthropic products for the next six months.

Meta has published an updated Advanced AI Scaling Framework and previewed a Safety & Preparedness Report for its Muse Spark model. The framework broadens risk evaluations to include chemical, biological, cybersecurity, and loss-of-control risks, and applies across open, API, and closed deployments. For Muse Spark, Meta tested the model before and after applying safeguards against thousands of adversarial scenarios, confirming it lacks autonomous capability sufficient to pose control risks and shows strong ideological balance in responses.

Google Colab has added two new features to its Gemini AI integration: Learn Mode and Custom Instructions. Learn Mode turns Gemini into a personal coding tutor, providing step-by-step guidance and concept explanations instead of simply generating copy-paste code. Custom Instructions let users tailor Gemini's behavior to specific coding styles, class syllabi, or preferred libraries, and are stored at the notebook level so they travel with shared notebooks, giving collaborators the same personalized AI experience.

Factory.ai has launched a native desktop app for its AI Droids, available now on macOS and Windows across all plans. The app supports multi-agent sessions, letting users run several Droids simultaneously with separate context and history. Key features include Droid Computers with persistent storage, BYO machine registration, local model support via Ollama or vLLM, computer use capabilities, and VS Code integration. Enterprise data shows users running both CLI and desktop interfaces complete 4.6x more sessions than CLI-only users.

Anthropic has redesigned its Claude Managed Agents hosted service by decoupling three core components: the harness (brain), sandbox (hands), and session log. Previously coupled in a single container, the architecture caused debugging difficulties and security risks. The new design lets each component fail or be replaced independently. A harness crash is recovered via wake(sessionId), while sandboxes become disposable cattle. The restructure cut median time-to-first-token by 60% and p95 latency by over 90%, and eliminated credential exposure in sandboxes.

A hacker using the name FlamingChina allegedly breached China's National Supercomputing Center in Tianjin, claiming to have stolen over 10 petabytes of sensitive data including classified defense documents, missile schematics, and aerospace research. The attacker reportedly gained access via a compromised VPN and used a botnet to extract data over six months undetected. Cybersecurity experts at SentinelOne who reviewed samples called them credible. Full dataset access is priced at hundreds of thousands of dollars in cryptocurrency.

OpenAI has released a Child Safety Blueprint, a policy framework aimed at combating AI-enabled child sexual exploitation. The blueprint focuses on three priorities: modernizing laws to address AI-generated CSAM, improving provider reporting and coordination with law enforcement, and embedding safety-by-design measures into AI systems. Developed with partners including NCMEC, Thorn, and the Attorney General Alliance—co-chaired by North Carolina AG Jeff Jackson and Utah AG Derek Brown—the framework targets prevention before harm occurs.

Cursor has launched a new feature allowing users to run AI coding agents on any machine and control them remotely from a phone or other device. Developers can now kick off agents from their phone to run on a remote development box, enabling asynchronous, location-independent workflows. This means coding tasks can be delegated and executed on powerful remote machines without requiring the user to be physically present at their workstation.

HeyGen has launched Avatar V, a new AI avatar tool that captures a user's identity in just 15 seconds and maintains character consistency across all generated videos. Users can then modify the look, outfit, and setting to produce unlimited personalized video versions while preserving their core identity. The feature addresses a longstanding challenge in AI video generation: keeping a consistent character appearance across different scenes and styles.