AI News Today December 2025: The Month the Rulebooks Collided

Anyone who searched for ai news today december 2025 during that month met a moving target. No single story held the headline for long. A pattern did emerge instead: governments on three continents moved on AI rules in the same weeks that the biggest labs shipped new models and signed record deals. This page pulls the month back together, regulation first, and every story carries the date it happened and the source that announced or reported it.

The short version runs like this. On December 11, the White House ordered federal agencies to challenge state AI laws. Eight days later, the governor of New York signed a state AI safety law in open defiance of that push. In Europe, lawmakers spent the month weighing a plan to push the AI Act’s toughest deadlines back by more than a year. Around those fights, OpenAI shipped GPT-5.2, Google made Gemini 3 Flash the default model in its app and Search, Disney put $1 billion and more than 200 characters behind OpenAI, Microsoft pledged $17.5 billion to India, and Apple replaced the executive in charge of AI.

Washington moves first: the order against state AI laws

The us ai regulation news today december 2025 searches spiked on one date, December 11, and for good reason. That day President Donald Trump signed Executive Order 14365, titled Ensuring a National Policy Framework for Artificial Intelligence. The order argues that a patchwork of 50 state regulatory regimes makes compliance harder, especially for startups, and that the United States needs one minimally burdensome national standard instead.

The order does not write that standard. It aims the federal government at the state laws that already exist. It tells the Justice Department to stand up an AI Litigation Task Force within 30 days to challenge state laws in court. It tells the Commerce Department to publish an evaluation of state AI laws within 90 days, naming the ones it considers onerous. It links federal broadband money, the BEAD program, to that list: states with onerous AI laws can lose eligibility for remaining funds. And it sends the FCC toward a single federal reporting and disclosure standard for AI models, and the FTC toward a policy statement on when state laws that force changes to truthful AI outputs must give way.

Two limits matter, and honest coverage kept both in view. First, an executive order cannot repeal a state law. The state laws stay in force unless courts strike them down or Congress replaces them, which is why legal analysts described the order as the start of a fight, not the end of one. The law firm Crowell & Moring published a plain breakdown of the order that walks through each lever. Second, Congress had its own chance that month and passed: lawmakers removed a provision from the annual defense bill that would have stopped states from enforcing their AI laws.

For companies, the practical effect landed at once. A business that sells AI in California, Colorado, Texas and New York now tracks state duties that Washington says it wants gone, with no date for when a court or Congress settles the question. That is why so many firms treat a live inventory of their own AI systems as legal protection, a habit our guide to AI governance in a business context sets out in full.

New York answers back: the RAISE Act

Eight days after the order, on December 19, Governor Kathy Hochul signed the Responsible AI Safety and Education Act, known as the RAISE Act. It became the first state AI law signed after the White House move, and its target is narrow and large at once: the biggest developers of frontier models, the most powerful AI systems, that operate in New York.

The law asks those developers to publish safety frameworks, put security protocols in writing, and report critical safety incidents to the state. It creates an oversight office inside the Department of Financial Services, carries civil penalties that the agreed final text sets at up to $1 million for a first violation and $3 million for repeat ones, and is due to take effect on January 1, 2027. Hochul signed it with a deal attached: lawmakers agreed to pass follow-up changes in January 2026 that align the law with California’s Transparency in Frontier Artificial Intelligence Act, including a revenue line of $500 million a year for the developers it covers.

Read together, December 11 and December 19 told the real US story of the month. The federal government said state AI rules threaten national leadership. New York signed one anyway, modeled on California’s. Until a court ruling or a federal law settles which side wins, both rulebooks exist at once, and the state one has dates on it.

Europe debates a pause in the AI Act

The eu ai act news today december 2025 had a different flavor: process, not verdicts. The AI Act, the European Union’s flagship AI law, took effect in August 2024 with its duties phased in over time. Its strictest part, the rules for high-risk AI systems such as tools used in hiring, credit and policing, is set to apply from August 2, 2026. On November 19, 2025, the European Commission proposed the Digital Omnibus on AI, a package that would link the start of those high-risk duties to the readiness of the standards and guidance companies need, with a backstop date of December 2, 2027 at the latest.

Through December, the proposal sat with the European Parliament’s internal market and civil liberties committees, the two committees that share the file. Industry groups such as the CCIA called the delay necessary, arguing that companies cannot comply with rules whose measuring tools do not exist yet. Consumer groups such as BEUC argued the package went far beyond a tidy delay and weakened duties that should stay. Nothing was adopted in December, and that was the news: the world’s most complete AI law spent the month openly negotiating with its own deadlines. Law firm DLA Piper keeps a running analysis of the omnibus that tracks the proposal clause by clause.

For US readers, the contrast is the story. Washington tried to slow state rules from outside, by pressure. Brussels worked from inside its own law, by amendment. Both sides used the same argument, that rules written before the tools to measure them can do harm, and both left December with the argument unresolved.

The models still shipped: GPT-5.2 and Gemini 3 Flash

Strip the rule fights away and the ai news today december 2025 still had a model race at its center. Google had launched Gemini 3 in November to strong reviews, and reporting in the first days of December said OpenAI chief executive Sam Altman had declared a code red inside the company, pausing side projects to push ChatGPT’s speed, reliability and personalization. On December 11, the answer shipped: GPT-5.2, in Instant, Thinking and Pro versions.

OpenAI aimed the release squarely at professional knowledge work: spreadsheets, presentations, code, images and long documents. The company said GPT-5.2 Thinking beats or ties human professionals on 70.9 percent of comparisons in GDPval, its test of real work tasks across 44 occupations. Treat that as a company claim from a company test, not an independent verdict. The rollout began in ChatGPT for paid plans, and the model reached developers in the API the same day, with GPT-5.1 kept on as a legacy option for three months.

Google answered on December 17 with Gemini 3 Flash, a faster member of the Gemini 3 family, and made the bolder distribution move: the model became the default in the Gemini app and in AI Mode in Search worldwide, replacing Gemini 2.5 Flash for everyday users. Google confirmed the rollout in its Workspace updates two days later. Google published a 90.4 percent score for the model on GPQA Diamond, a test of hard science questions, and sold the release on one phrase: frontier intelligence built for speed.

Disney bets $1 billion on OpenAI

The third story of December 11 belonged to Hollywood. The Walt Disney Company announced a three-year licensing agreement with OpenAI and a $1 billion equity investment in the company, with warrants to buy more. Disney became the first major studio to license its characters into a generative AI platform on purpose: from early 2026, fans can prompt Sora, OpenAI’s video generator, to make short clips with more than 200 characters from Disney, Pixar, Marvel and Star Wars, and use the same cast in ChatGPT Images.

The boundaries matter as much as the headline. The deal covers animated, masked and creature characters, plus costumes, props, vehicles and settings. It does not include the likenesses or voices of human actors. Disney also signed on as a major OpenAI customer, building products on its APIs and rolling ChatGPT out to employees, and a curated set of fan videos is planned for Disney Plus. A studio that had spent much of 2025 fighting AI companies over copyright chose, in December, to sell a license instead, and every other studio now has a price and a template to react to.

The deal also closed a loop with the autumn. Our recap of the AI news of October 12, 2025 showed an industry sprinting toward agents and video tools. December showed the money arriving behind the video tools, with the world’s most guarded character catalog attached.

The money and the management: Microsoft and Apple

The business pages of ai news today december 2025 told the same story at a larger scale: AI had become an infrastructure budget and a leadership question. On December 9, Microsoft chief executive Satya Nadella met Prime Minister Narendra Modi in New Delhi and announced a $17.5 billion investment in India across 2026 to 2029, the company’s largest commitment in Asia. The money expands cloud and AI infrastructure, funds skills programs and builds sovereign cloud options, and it sits on top of a $3 billion pledge Microsoft made earlier in 2025.

The centerpiece is a new India South Central cloud region in Hyderabad, due to go live in mid-2026. The logic matches the regulation sections above: countries increasingly want AI capacity, and the data behind it, inside their own borders, and the company that builds the local data centers gets to sell the local cloud.

Apple made its December move on the first of the month, and it was about people, not products. In an Apple Newsroom announcement dated December 1, the company said John Giannandrea, its senior vice president for Machine Learning and AI Strategy since 2018, would step down, advise the company, and retire in the spring of 2026. Amar Subramanya joined as vice president of AI, reporting to software chief Craig Federighi, and took charge of Apple Foundation Models, machine learning research, and AI Safety and Evaluation. The rest of Giannandrea’s organization moved to operations chief Sabih Khan and services chief Eddy Cue.

Subramanya’s resume explains the hire. He spent 16 years at Google, where he led engineering for the Gemini Assistant, then served as a corporate vice president of AI at Microsoft. Apple, whose Apple Intelligence features arrived later and drew cooler reviews than its rivals’ launches, hired the person who shipped the rival assistant. His first test is the more personal Siri that Apple has promised for 2026.

What December 2025 set up for 2026

Read as one file, the ai news today december 2025 set four clocks ticking. The Justice Department task force was due within 30 days of December 11, and the Commerce Department’s list of onerous state laws within 90, which puts both landings in the first quarter of 2026. The EU has to settle the omnibus before August 2, 2026, or the original high-risk dates apply by default. New York’s RAISE Act duties land on January 1, 2027, a fixed target the big labs can plan against. And Disney’s licensed characters are due to reach Sora in early 2026, the first real test of whether fans and studios can share a generative video tool.

Two slower stories run underneath all four. Platform enforcement kept tightening through the year, a theme our explainer on why Google deletes apps from the Play Store follows from the app store side. And as AI answers became the front door to the news itself, publishers kept asking what generative engine optimization is, because a recap like this one now competes with the summary an AI gives before the click.

Conclusion

December 2025 did not produce one AI story. It produced the shape of the next few years: a federal government pushing back on state rules, states signing them anyway, Europe renegotiating its own law from inside, and labs and studios spending billions as if none of it slowed them down. That is the honest summary of ai news today december 2025: the rules and the releases now land in the same weeks, and neither side waits for the other anymore.