Microsoft Research presents Lens, a text-to-image model with just 3.8 billion parameters that matches much larger rivals on benchmarks, at a fraction of the training cost. The secret sauce: 800 million detailed image captions generated by GPT-4.1 instead of vague web alt-text. Code and weights are openly available under an open-source license.

Google has ordered more than three million AI chips from Intel for 2028. Nvidia is testing Intel's manufacturing tech for its upcoming Feynman architecture. Both moves come as TSMC can't keep up with AI chip demand. Intel's long-struggling foundry division is getting a rare second chance.

Only 26 percent of companies have full visibility into their AI costs, a KPMG survey finds.

OpenAI is planning the biggest overhaul of ChatGPT since its launch. The chatbot will become a "superapp" bundling coding tools, AI agents, and partner apps like Canva and Booking.com. "Chat is dead," the company says internally. The future supposedly belongs to agents that handle tasks on their own.

Perplexity's new "Search as Code" architecture dumps rigid search APIs and lets AI models write their own search routines in Python. By letting the agent handle its own filtering and deduplication inside a sandbox, the system beats OpenAI and Anthropic on key benchmarks, while cutting token costs by up to 85 percent.

OpenAI's new Lockdown Mode for ChatGPT disables web access, Deep Research, and Agent Mode to make data theft through prompt injection attacks harder. The mode doesn't fully prevent such attacks, it only blocks the final step in an exfiltration chain. Prompt injection remains an unsolved problem.

SpaceX is leasing AI computing capacity to Google for $920 million per month, according to an SEC filing. The deal gives Google access to about 110,000 Nvidia chips to meet demand for its Gemini Enterprise platform. That one of the world's largest cloud providers needs to rent capacity externally shows how scarce AI infrastructure has become, and how tightly big tech companies' businesses are now intertwined.

Meta is developing a paid AI agent product called "Hatch" that could cost up to $200 per month. Users describe what they need in simple language, and Hatch builds working tools, schedules appointments, or sends emails. CEO Mark Zuckerberg sees the product as a way to open up new revenue streams beyond advertising and refinance the company's massive AI investments.

Alibaba's Qwen team has released Qwen3.7-Plus, a multimodal agent model that combines visual perception, GUI operation, and coding in a single agent loop. In a demo, an agent built on the model autonomously developed a vocabulary learning app, producing over 10,000 lines of code across 1,000 agent calls over eleven hours. The model leads on-screen understanding in Qwen's own benchmarks, but overall performance is mixed. Qwen3.7-Plus is a proprietary offering with no open weights, priced well below Western frontier models.

Florida is the first US state to sue OpenAI and CEO Sam Altman personally over risks to minors, missing age checks, and inadequate safety investment. The 83-page complaint treats ChatGPT as a product subject to liability and threatens billions in penalties. The legal approach could set a precedent for the entire chatbot industry.

Anthropic is sharing internal data showing how much Claude is speeding up its own AI development: more than 80 percent of production code now comes from Claude, and engineers are shipping eight times as much code per day as in 2024. The goal is AI that improves itself, which would trigger a massive acceleration. This is why Anthropic is now pushing for the option of a verifiable, global development pause. The company says it would stop if other frontier labs demonstrably do the same.

Artificial Intelligence is changing the world. THE DECODER brings you all the news about AI.

For the first time, Google is giving website operators an opt-out toggle in Search Console for AI search features like AI Overviews and AI Mode, which together already reach more than 3.5 billion monthly users. New performance reports break out impressions separately. The move was prompted by the UK's Competition and Markets Authority (CMA), which sees website operators at a severe disadvantage.

Google Deepmind's Gemma 4 12B is an open-source model that processes text, images, and audio natively and runs on laptops with just 16 GB of RAM. It nearly matches the twice-as-large 26B model in benchmarks and ships under an Apache 2.0 license for commercial use.

Sam Altman, Dario Amodei, Demis Hassabis, and other tech leaders are urging the US government to make screening of synthetic DNA orders a legal requirement. AI systems already outperform PhD-level virologists on lab procedures, the signatories warn, raising the risk of misuse for biological weapons.

The White House has issued an executive order requiring agencies like the Pentagon and CISA to strengthen cyber defense with AI tools within 30 days. AI developers can voluntarily submit models for security testing, but the order explicitly rules out mandatory approval. Given recent government pressure on AI companies, how voluntary this cooperation really is remains an open question.

For the first time, Google is giving website operators an opt-out toggle in Search Console for AI search features like AI Overviews and AI Mode, which together already reach more than 3.5 billion monthly users. New performance reports break out impressions separately. The move was prompted by the UK's Competition and Markets Authority (CMA), which sees website operators at a severe disadvantage.

Google Deepmind's Gemma 4 12B is an open-source model that processes text, images, and audio natively and runs on laptops with just 16 GB of RAM. It nearly matches the twice-as-large 26B model in benchmarks and ships under an Apache 2.0 license for commercial use.

Hackers took over prominent Instagram accounts, including the Obama White House page, by simply asking Meta's AI support chatbot to change the email address on file. Two-factor authentication was bypassed entirely. Meta has patched the flaw, but security researchers say another exploit is already circulating on Telegram.

Anthropic is scaling up Project Glasswing with 150 new partners across more than 15 countries, all using Claude Mythos Preview to scan critical infrastructure for security flaws. Partners already on board have found over 10,000 serious vulnerabilities. At the same time, Anthropic is selling a commercial fix with Claude Security, profiting from both sides of the problem.

OpenAI is expanding Codex with role-specific plugins for data analysis, sales, and investment banking. Five million people use the tool each week, and one in five isn't a developer, the company says. That non-developer group is growing three times faster than the developer base, a sign that OpenAI is positioning Codex as an all-purpose work app.

Anthropic has confidentially filed a draft IPO registration with the US Securities and Exchange Commission (SEC). The company behind the chatbot Claude is valued at just under $1 trillion after its latest funding round. Competitor OpenAI is also gearing up for an IPO. The race for investor dollars in the AI sector is heating up.

Nvidia used GTC Taipei to launch a series of models for robots, autonomous vehicles, and video systems. The centerpieces are the new world model Cosmos 3, a significantly scaled-up driving model called Alpamayo 2 Super, and an open reference platform for humanoid robots.

Chinese AI company MiniMax has released its new model M3. It's billed as the first open-weight model to combine top-tier coding performance, a one-million-token context window, and native multimodality.

SoftBank plans to build AI data centers with up to 5 gigawatts of capacity in France, the company's largest AI infrastructure investment in Europe, at up to 75 billion euros. By 2031, facilities worth 45 billion euros are set to go up at three sites in northern France. SoftBank's mega announcements keep stacking up worldwide, but many projects have yet to materialize.

Nvidia is pushing into the PC market with its own chips as the main processor. The first Windows computers from Dell and Microsoft's Surface line are set to be unveiled next week at Computex and Build. Microsoft is also planning new software likely based on the OpenClaw framework that lets AI agents handle tasks locally on Windows PCs, a second shot after the Copilot+ PC concept largely flopped.

Anthropic bans AI during job interviews and runs candidates through up to five rounds testing skills, values, and ethical thinking. Salaries go up to $850,000, and some applicants pay $4,600 for prep coaching run anonymously by current AI company employees.

Nvidia is pushing into the PC market with its own chips as the main processor. The first Windows computers from Dell and Microsoft's Surface line are set to be unveiled next week at Computex and Build. Microsoft is also planning new software likely based on the OpenClaw framework that lets AI agents handle tasks locally on Windows PCs, a second shot after the Copilot+ PC concept largely flopped.

OpenAI is updating GPT-5.5 Instant for more natural responses and dropping the Canvas feature from its latest models. Writing and coding tasks will run directly in the chat instead. The company is also retiring the older o3 and GPT-4.5 models from ChatGPT, with both shutting down by August 2026 at the latest.

OpenAI's Codex app now runs on Windows 11 with "Computer Use": the AI can independently control programs, test apps, and hunt for bugs. When no one's at the PC, the ChatGPT mobile app lets users start and monitor tasks remotely from their phone.

Anthropic raises $65 billion in a Series H round at a $965 billion valuation. Annualized revenue tops $47 billion, according to CFO Krishna Rao. The company plans to invest in safety research, computing capacity, and expanding its Claude product lineup.

OpenAI is offering its life sciences model GPT-Rosalind for free through the new Rosalind Biodefense program, aimed at pandemic preparedness and biodefense. Early partners include Lawrence Livermore National Laboratory, Johns Hopkins, and vaccine initiative CEPI. Applications are open worldwide.

OpenAI is updating GPT-5.5 Instant for more natural responses and dropping the Canvas feature from its latest models. Writing and coding tasks will run directly in the chat instead. The company is also retiring the older o3 and GPT-4.5 models from ChatGPT, with both shutting down by August 2026 at the latest.

Google Cloud has unveiled "AI Threat Defense," a platform designed to automatically find, assess, and patch security flaws in enterprise systems. The company bundles technologies it partly acquired through acquisitions.

Amazon MGM Studios and AWS are launching a "GenAI Creators' Fund" that gives filmmakers money and access to the in-house AI platform "Project Nara." Three animated series are already in production - the teams had five weeks for their pilots. Amazon says it now has the "only end-to-end AI content ecosystem in the industry."

Meta is rolling out paid add-ons for Instagram, Facebook, and WhatsApp worldwide while building a separate paid AI offering.

China's police are upgrading millions of old surveillance cameras with AI. Manufacturers like Hikvision and Huawei now ship cameras with built-in computer vision and language models that automatically detect crowds, suspicious behavior, or unauthorized access. Instead of reviewing footage manually, officers just type a text query. Human Rights Watch warns this creates unprecedented behavioral surveillance at scale.

Robinhood now lets customers connect AI agents like Anthropic's Claude to a separate investment account via MCP. The agents can trade stocks on their own. US brokerage regulator FINRA already flags such agents as a new risk area, warning about unchecked decisions. Robinhood itself admits the product isn't for everyone.

YouTube is tightening its AI labeling rules. Labels for photorealistic or heavily AI-altered content will now show up in more visible spots, below the player for long videos and as an overlay on Shorts. Starting May 2026, an automatic detection system will flag AI-generated content even if creators don't disclose it. Recommendations and monetization won't be affected.

For agentic AI to deliver material benefits to organizations, it can’t be layered onto existing operations. Instead, enterprise leaders must approach it as a systems-level change.

What do the numbers really say about the impact of artificial intelligence on the labor market? The answer might surprise you.

Shortly after OpenAI disproved Erdős' unit-distance conjecture, Anthropic shows Claude Mythos can solve the problem too - "over the weekend." Engineer Sholto Douglas says Mythos cracked the 1946 conjecture with a "cute, simple proof," a sign of "serious overhang" in AI-driven math discoveries.

Google Deepmind's AlphaProof Nexus has autonomously solved nine open Erdős problems, including two that stumped mathematicians for 56 years, for just a few hundred dollars per problem in inference costs. Unlike OpenAI's natural-language approach, the system uses the Lean compiler to verify every proof step automatically. Still, the overall success rate sits at just 2.5 percent.

Anthropic co-founder Christopher Olah was invited to speak at the launch of Pope Leo XIV's encyclical "Magnifica Humanitas" and used the stage to claim AI models show evidence of introspection and emotion-like states. The Pope's own document struck a different tone: "These systems merely imitate certain functions of human intelligence."

GenAI navigation on shaky ground.

Researchers from UMD, Google, Meta, and other institutions use AutoTTS to let a coding agent independently discover control algorithms for AI reasoning. The algorithm it found cuts compute by about 70 percent compared to standard self-consistency while matching its accuracy. The whole search cost $40 and took 160 minutes.

ByteDance Seed shows that a 7B model can answer questions on long, image-heavy documents more reliably than much larger models, even when documents are four times longer than anything it saw during training. Instead of transcribing pages, the model learns by answering questions and finding the right passages on its own.

Anthropic will likely keep supplying AI models to the NSA despite being labeled a "supply chain risk." Intelligence agencies lack Nvidia's latest Grace Blackwell chips, and Anthropic's "Mythos" model reportedly runs on older hardware too. The controversial "any lawful use" clause that derailed earlier talks is not part of the deal.

Anthropic's AI model Claude Mythos Preview, working with about 50 partners as part of Project Glasswing, has found over 10,000 critical vulnerabilities in system-critical software. The bugs are piling up faster than anyone can patch them. Anthropic warns this creates a high-risk transition period and says no company, itself included, has built safeguards strong enough to prevent misuse of these models.

Alibaba's Qwen team releases Qwen3.7-Max, a proprietary model built for long-running autonomous agent tasks. It matches Claude Opus 4.6 on benchmarks and beats Chinese rivals like DeepSeek V4 Pro and Kimi K2.6. The team also demos the model steering a four-legged robot.

OpenAI brings ChatGPT directly into PowerPoint. A new beta plugin creates presentations from notes, documents, or images and edits existing slides. The add-in is available worldwide across all tiers. OpenAI recommends saving important decks before using it.

Two years ago, an AI tool won Google DeepMind a Nobel. Researchers are now climbing toward a new goal.

OpenAI brings ChatGPT directly into PowerPoint. A new beta plugin creates presentations from notes, documents, or images and edits existing slides. The add-in is available worldwide across all tiers. OpenAI recommends saving important decks before using it.

With the new Codex feature Appshots, Mac users can send the contents of any app window to OpenAI's coding assistant Codex at the press of a button, giving it the context it needs for a task.

A reasoning model from OpenAI has disproved a conjecture by mathematician Paul Erdős on unit-distance geometry that stood open since 1946 - using tools from algebraic number theory that experts never expected in this context. Fields Medalist Tim Gowers calls the result "a milestone in AI mathematics" and warns: "We have still probably entered an era where it will become very difficult for humans to compete with AI at solving mathematical problems."

Anthropic is approaching its first profitable quarter, the Wall Street Journal reports - with a projected operating profit of $559 million on $10.9 billion in Q2 revenue. Just last summer, the company didn't expect to turn a profit before 2028. The main drivers are coding tools and agentic Claude usage, which at times pushed demand beyond available compute capacity.

US Cyber Command has launched a task force to run AI models from OpenAI, Google, and others on the most classified Pentagon and NSA networks. The trigger: AI systems like Anthropic's Claude Mythos can find security vulnerabilities faster than the best human hackers - and according to Anthropic, comparable tools could be widely available within six to 24 months.

Watch subscriber-only discussion going behind the scenes of the trial and the implications for the AI race.

Google Deepmind connects its Genie 3 world model to Street View imagery: users drop a pin on a map and get a walkable, AI-generated world based on a real place. Google's Street View data, collected over many years, becomes a strategic training resource for creative demos, but above all for AI agents and robots.

Stability AI has unveiled Stable Audio 3.0, a new generation of audio models - three of which ship with open weights. The models generate music tracks up to six minutes long and were trained entirely on licensed data, according to the company.

Google used its I/O developer conference to unveil a wave of new AI products. The highlights: a new model called Gemini 3.5 Flash, a multimodal model called Gemini Omni, and a personal agent named Gemini Spark that runs around the clock in the cloud. The Gemini app also gets a major refresh.

Google is restructuring its AI subscriptions at I/O 2026: three tiers from $7.99 to $99.99 per month with staggered usage limits, new models like Gemini Omni, and the AI agent Gemini Spark. Instead of daily prompt limits, Google is moving to a consumption-based compute model, a trend that's gaining traction across the industry.

After three weeks of dueling testimony, the jury decided Musk had sued the AI giant too late.

Elon Musk has lost his lawsuit against Sam Altman and OpenAI. The jury in Oakland dismissed the case after just two hours of deliberation. Musk had sought up to $134 billion. The judge said she would have been ready to dismiss the case "immediately." Musk's attorney reserved the right to appeal.

It’s been a year since the duo entered the US Army’s troubled augmented-reality contest. Here’s what it looks like so far.

The company has fallen behind its closest competitors where it matters most. Can it catch up?

OpenAI is merging ChatGPT, its coding agent Codex, and the developer API into a single product team led by Codex boss Thibault Sottiaux. The goal: a "super app" that also integrates the Atlas browser. Co-founder Greg Brockman officially takes over product strategy.

Oppo's Multi-X team released X-OmniClaw, an open-source agent that runs directly on Android devices and combines camera, screen, and voice to handle tasks in real apps. Instead of relying on cloud copies of the phone, the system uses local sensors; cloud compute only kicks in for reasoning. Tap paths get cloned as reusable skills, so the agent can jump straight to deeply nested app pages via deeplink next time around.

A consortium of 64 mathematicians built SOOHAK, a new AI benchmark with 439 handwritten tasks, including 99 that are deliberately unsolvable. Google's Gemini 3 Pro leads on research-level problems at 30 percent. But no model cracks 50 percent on spotting broken tasks. More compute makes models better at solving. It doesn't improve them at admitting a problem has no answer. SOOHAK tries to pin down the gap between a few flashy results and the broad research skills AI systems still lack.

The trial spilled plenty of dirt—and raised more questions than answers about how the AI giant should be governed.

Researchers at Carnegie Mellon University built a new benchmark that measures how far AI agents can go when exploiting real vulnerabilities in Google's V8 engine. Mythos leads GPT-5.5 by a wide margin but costs twelve times as much.

OpenAI has acquired Weights.gg, a small startup that let users create and share AI voice clones of celebrities like Taylor Swift and Donald Trump. The team of around six now works at OpenAI, but the company doesn't plan to release a standalone cloning product.

Anthropic is raising another $30 billion just three months after a round of the same size. The AI lab's valuation jumps to $900 billion, surpassing rival OpenAI for the first time. Fueling the surge: annualized revenue approaching $45 billion, a fivefold increase since the end of 2024.

OpenAI is turning ChatGPT into a personal financial assistant. Pro users in the US can now connect their bank accounts through Plaid to get personalized analysis based on real transaction data. The feature runs on GPT-5.5 Thinking and will eventually roll out to all users. OpenAI warns the chatbot is still not a licensed financial advisor.

Arxiv, the influential preprint server where researchers worldwide publish their work before formal peer review, is tightening its rules on AI-generated content.

Microsoft has built MDASH, a system that pits more than 100 specialized AI agents against each other to find software vulnerabilities. On Patch Tuesday alone, the system uncovered 16 security flaws in Windows, four of them critical. Microsoft isn't saying which AI models power the system.

The US has cleared roughly ten Chinese companies—including Alibaba, Tencent, and ByteDance—to buy up to 75,000 Nvidia H200 chips each. But not a single chip has shipped. According to Commerce Secretary Lutnick, Beijing is blocking the purchases to protect its domestic chip industry.

People report that their personal contact info was surfaced by Google AI—and there’s apparently no easy way to prevent it.

People report that their personal contact info was surfaced by Google AI—and there’s apparently no easy way to prevent it.

Meta is rolling out "Incognito Chat" for Meta AI on WhatsApp and in the Meta AI app. According to Mark Zuckerberg, conversations are processed in a protected server environment that even Meta can't access, and chat histories disappear when the session ends. Zuckerberg claims Meta is the first AI lab to offer this level of private AI usage.

Anthropic now leads OpenAI in B2B adoption for the first time, with 34.4 percent of US companies on the Ramp AI Index compared to OpenAI's 32.3 percent. Anthropic quadrupled its reach in just one year, but three factors could erode that lead quickly.

Google's Threat Intelligence Group has identified the first known case of an attacker using AI to discover and weaponize a zero-day vulnerability. Google says it stopped the planned mass attack. State-backed actors from China, North Korea, and Russia are also using AI to find vulnerabilities and disguise malware code.

Isomorphic Labs, the AI drug research company led by DeepMind co-founder Demis Hassabis, has closed a $2.1 billion Series B round led by Thrive Capital. The funding will go toward expanding its in-house platform IsoDDE and advancing drug candidates toward clinical trials.

With Gemini Intelligence, Google is introducing new AI features for Android that automate multi-step tasks, summarize web content, fill out forms, and turn spoken thoughts into polished text messages.

OpenAI is facing a lawsuit over the mass shooting at Florida State University. According to the complaint, the shooter spent months talking to ChatGPT about guns and shootings. Florida's attorney general has launched a criminal investigation, saying: "If ChatGPT were a person, it would be facing charges for murder." The case adds to a growing wave of lawsuits targeting AI chatbots.

OpenAI has offered the EU Commission direct access to its new GPT-5.5 Cyber model for security review, with talks already underway. Anthropic is proving harder to pin down: after four to five meetings on its Mythos model, regulators still don't have access. The gap highlights how dependent Europe's AI oversight remains on voluntary cooperation from the companies it aims to regulate.

Language models find security flaws faster and turn patches into working exploits in minutes. A veteran researcher says the established disclosure process needs to change.

Palisade Research shows that AI agents can hack remote computers, copy themselves onto them, and form replication chains. In one year, the success rate jumped from 6 to 81 percent. The researchers expect remaining barriers to fall as models get better at hacking.

METR can barely measure Claude Mythos Preview with its current test suite. Only five out of 228 tasks cover the relevant capability range. Meanwhile, Palo Alto Networks reports that frontier models autonomously chain vulnerabilities, shrinking the time from initial access to data exfiltration to just 25 minutes. Evaluation methods are growing more slowly than the models themselves, and that may be the bigger problem.

ByteDance is raising its planned AI spending for 2026 to over 200 billion yuan (roughly $30 billion), at least a 25 percent jump from earlier plans. The TikTok parent is increasingly turning to Chinese chips. Still, the figure looks modest next to the $725 billion that Google, Amazon, Microsoft, and Meta are planning to spend combined.

OpenAI president Greg Brockman said Elon Musk wanted the company to create a for-profit entity—and endured a public peek into his diary.

Fields Medalist Timothy Gowers had ChatGPT 5.5 Pro tackle open problems in number theory. The model improved an exponential bound to a polynomial one in under an hour. An MIT researcher involved calls the key idea "completely original." Gowers' takeaway: the bar for mathematical contributions is now proving something LLMs can't.

OpenAI's custom AI chip project with Broadcom has hit a funding wall. Broadcom won't finance production unless Microsoft commits to buying 40 percent of the chips, and Microsoft hasn't agreed yet. OpenAI manager Sachin Katti called the dependency "financially unattractive" in an internal message. The first phase alone costs around 18 billion dollars.

According to the Financial Times, Anthropic's planned funding round is taking shape. The round aims to raise up to $50 billion, which would value the company at roughly $900 billion.

Anthropic's Natural Language Autoencoders make Claude Opus 4.6's internal activations readable as plain text. Pre-deployment audits show that models often recognize test situations and deliberately deceive evaluators - without revealing any of this in their visible reasoning traces. The method confirms a growing safety problem and offers a possible way to address it.

OpenAI is releasing GPT-5.5-Cyber, a model variant that rejects far fewer security requests and even actively executes exploits against test servers. Access is limited to verified defenders of critical infrastructure, including partners like Cisco, CrowdStrike, and Cloudflare. The model competes directly with Anthropic's Mythos Preview.

OpenAI is shipping three new voice models—GPT-Realtime-2, GPT-Realtime-Translate, and GPT-Realtime-Whisper—that can reason in real time, translate across 70+ languages, and transcribe live speech. GPT-Realtime-2 brings reasoning that OpenAI says matches GPT-5.

The EU has agreed on simplified AI rules. The "Digital Omnibus on AI" pushes back deadlines for high-risk AI to late 2027 or 2028 and eases requirements for small and medium-sized businesses. "Nudification" apps are now explicitly banned. The labeling requirement for deepfakes and AI-generated text still takes effect in August 2026.

Anthropic is set to tap into Elon Musk's Colossus 1 supercomputer. Behind the surprise deal lie a compute crunch, a looming IPO, and a remarkable about-face by Musk.

According to a report by The Information, Anthropic has committed to spending roughly $200 billion on Google Cloud over the next five years - more than 40 percent of Google's entire cloud backlog. Together with OpenAI, the two money-losing startups account for roughly half of the $2 trillion in committed cloud revenue at Amazon, Microsoft, Google, and Oracle. Whether the 20- to 30-fold revenue growth both companies are projecting by 2029 will justify these numbers remains an open question.

Anthropic is taking over the full computing capacity of SpaceX's Colossus 1 data center, more than 300 megawatts and over 220,000 NVIDIA GPUs, expected to come online within a month. The company is also doubling rate limits for Claude Code and significantly raising API limits for Opus models.

OpenAI is swapping out ChatGPT's default model for GPT-5.5 Instant. In internal testing, the update produced 52.5 percent fewer hallucinated claims on high-risk topics like medicine and law. A new feature called "memory sources" lets users see which stored context shaped a given response. The model is rolling out to all ChatGPT users right away, though personalization based on past chats, files, and Gmail launches first for Plus and Pro users on the web.

OpenAI is swapping out ChatGPT's default model for GPT-5.5 Instant. In internal testing, the update produced 52.5 percent fewer hallucinated claims on high-risk topics like medicine and law. A new feature called "memory sources" lets users see which stored context shaped a given response. The model is rolling out to all ChatGPT users right away, though personalization based on past chats, files, and Gmail launches first for Plus and Pro users on the web.

The US Department of Commerce is expanding its AI safety testing: Following Anthropic and OpenAI, Google Deepmind, Microsoft, and xAI have now signed agreements with the Center for AI Standards and Innovation. The companies provide models with reduced safety guardrails for testing in classified environments amid growing cybersecurity risks and an intensifying tech race with China.

Here’s what we learned from Elon Musk’s testimony, and what to expect this week.