Top AI News Today: Aug 31 (13 Stories)

AI News for August 31, 2026: A Music Lawsuit, a Broken Partnership, and a Price Hike

The last day of August brought AI news of the legal and financial kind more than the model release kind. Sony Music Publishing and Warner Chappell Music sued Anthropic on Friday, accusing it of pirating thousands of songs to train Claude, in a case that names CEO Dario Amodei personally. A day later, OpenAI announced it is cutting off the coding tool Cursor from its models after Elon Musk's SpaceX completed its acquisition of the company, citing a history of contract violations by Musk's other businesses.

On the pricing and policy side, Claude Sonnet 5's discounted launch pricing ends today, reverting to its standard rate, ChatGPT is retiring its standalone DALL-E tool, and California lawmakers are racing to pass a wave of AI bills before their legislative session adjourns tonight. Here are the 13 stories that matter most from today's AI news, explained in plain English.

Sony Music and Warner Chappell sue Anthropic over song lyrics

Sony Music Publishing and Warner Chappell Music filed a lawsuit against Anthropic on August 28, 2026 in federal court in Northern California, naming CEO Dario Amodei and co-founder Benjamin Mann as individual defendants alongside the company. The 48 page complaint accuses Anthropic of running a brazen campaign of illegally torrenting, scraping, and downloading copyrighted musical works to train and operate its Claude models.

The publishers say tens of thousands of songs were affected, naming titles including Eye of the Tiger, Uptown Funk, and Taylor Swift's Paper Rings, and claim Anthropic pulled lyrics from sites like Musixmatch and LyricFind alongside general datasets such as Common Crawl and Books3. They are asking for up to 150,000 dollars in damages per infringed work, which could add up to billions of dollars if the case succeeds, and they point to a separate ruling that already described one of Anthropic's earlier book piracy practices as straightforward piracy at massive scale.

This is not Anthropic's first music related lawsuit: Concord Music Group and Universal Music Group sued the company in January over a narrower set of about 500 compositions, and BMG has its own separate case. What makes this new suit notable is its scope, covering tens of thousands of works rather than hundreds, and its timing, landing just as Anthropic pursues a funding round reportedly valuing the company at close to 2 trillion dollars ahead of a possible public listing.

OpenAI cuts off Cursor's access to its models after the SpaceX takeover

OpenAI notified SpaceX on August 29, 2026 that it intends to wind down the contract supplying its AI models to the coding platform Cursor, proposing a cutoff date of November 12, 2026. SpaceX completed its roughly 60 billion dollar acquisition of Cursor on August 14, folding the coding tool into an AI portfolio built around its own Grok models.

OpenAI says it cannot be confident SpaceX will honor its usage terms, pointing to Elon Musk's testimony under oath earlier this year that xAI, another Musk owned company, had used a technique called distillation on OpenAI's models to help train its Grok chatbot in violation of OpenAI's terms of service. OpenAI also said its next model, code named Astra, will not be made available to Cursor at all, citing a new level of accountability over how that model gets deployed.

Cursor CEO Michael Truell said OpenAI's models make up only about 5 percent of Cursor's user traffic, and that the two companies are still talking. Anthropic, notably, has its own business relationship with SpaceX, leasing computing capacity from Musk's company even as Cursor now sits inside the same corporate family. The episode echoes Anthropic's own decision last year to cut off the coding tool Windsurf's access to Claude, showing that AI labs increasingly treat model access as a lever they are willing to pull when a customer's ownership or behavior changes.

Claude Sonnet 5's launch pricing ends today

Claude Sonnet 5's introductory pricing of 2 dollars per million input tokens and 10 dollars per million output tokens ends today, August 31, 2026, reverting to its standard rate of 3 dollars and 15 dollars per million tokens. Anthropic had offered the lower launch pricing since Sonnet 5's release on July 8, 2026.

A jump from 2 to 3 dollars per million input tokens might look small on paper, but for a business running Sonnet 5 across thousands of daily requests, a 50 percent price increase on both input and output tokens can meaningfully change a monthly bill overnight. Anthropic gave the change ahead of time in its documentation, part of a broader pattern this year of labs using cheap launch pricing to win developers over, then normalizing rates once a model is established.

This is one of several dated deadlines Anthropic has set for the Claude line this year, alongside retirement dates already passed for Claude Sonnet 4 and Opus 4 in June and Claude Opus 4.1 in August. Anthropic commits to giving at least 60 days of notice before retiring a model outright, though pricing changes like this one can arrive with less warning than a full retirement.

ChatGPT retires its standalone DALL-E image generator

ChatGPT is retiring its standalone DALL-E GPT on August 30, 2026, according to release notes OpenAI published in late July. The company is encouraging anyone who wants to keep images they created with the tool to download them before the shutdown.

The DALL-E GPT was a dedicated chat persona inside ChatGPT built specifically around OpenAI's older image model, separate from the general purpose ChatGPT Images feature that most people now use to generate pictures inside a normal conversation. Retiring the standalone version is a cleanup move rather than the end of image generation in ChatGPT, since ChatGPT Images continues to work exactly as before.

This kind of retirement has become routine as OpenAI folds older, single purpose tools into its main chat experience. A similar move happened on August 26, when OpenAI retired the older o3 reasoning model from ChatGPT's model picker, though it noted no

changes to o3's availability through the API. Anyone who built a habit or a workflow specifically around the DALL-E GPT persona will need to switch to asking ChatGPT directly for images instead.

California lawmakers race to pass a wave of AI bills before adjourning

California's legislature is scheduled to adjourn for the year at the end of business today, August 31, 2026, and lawmakers are racing to finish votes on roughly two dozen AI related bills still moving through the process. The bills cover chatbot safety, children's use of AI, medical and mental health disclosures, consumer rights, and oversight of the most powerful frontier models.

California passing AI laws matters well beyond the state's own borders, since many major AI companies are headquartered there and often end up building compliance for California's rules into their products everywhere, similar to how the state's emissions and privacy laws have shaped national policy in other industries before. Earlier in 2026, California and other states together passed 85 new AI related laws, and today's session is expected to add several more before the legislature breaks.

Among the bills still moving is a measure requiring companies deploying AI tools to make reasonable efforts to reduce biased outcomes and to monitor their systems for that risk over time, along with separate proposals on attorney use of AI and how public records rules apply to AI generated content. Because everything not signed by the time the session ends has to wait for next year, today's final votes will decide which of these rules actually reach the governor's desk in the near term.

Amazon will shut down Mechanical Turk after 21 years

Amazon Web Services told users it will shut down Mechanical Turk, its 21 year old task marketplace, on September 30, 2026, following an internal assessment of the platform's future. Mechanical Turk stopped accepting new customer sign ups ahead of the closure.

Launched in 2005, Mechanical Turk let businesses pay people small amounts of money to complete simple tasks that were easy for humans but hard for software at the time,

work Amazon founder Jeff Bezos once called artificial artificial intelligence. Ironically, that same kind of human labeled data became one of the key ingredients used to train the modern AI models that have now made much of that manual task work unnecessary.

Mechanical Turk's closure is a small but symbolic marker of how far AI has come in 21 years: a platform built specifically because computers could not yet do certain simple tasks is shutting down in the same year that AI agents are being handed increasingly complex, multi step jobs across finance, coding, and customer service with far less human involvement than Mechanical Turk ever required.

Marvell says its AI data center chip business will triple by 2028

Chipmaker Marvell said this week that its AI data center chip business is on track to reach 18 billion dollars in annual revenue by 2028, a forecast the company shared alongside other major tech earnings news this week. Marvell makes custom chips and networking components used inside the data centers that train and run AI models.

Unlike Nvidia, which sells complete general purpose AI chips to nearly every customer, Marvell focuses more on custom silicon built for specific large customers and on the networking hardware that connects thousands of chips together inside a data center. Growth forecasts like this one matter because they signal how much continued spending big technology companies expect to pour into AI infrastructure over the next two years, not just this one.

The forecast lands during the same week Nvidia posted record earnings of its own and reportedly began exploring a roughly 13 billion dollar acquisition of the open source AI hub Hugging Face, underscoring how much money continues to flow through the hardware side of the AI industry even as questions grow about whether some of that spending is more circular than it first appears. Investors have been watching networking and custom chip suppliers closely this year, since their order books tend to reveal how far ahead the largest cloud providers are actually planning their AI buildouts, regardless of which specific model happens to be topping the leaderboards that week.

OpenAI brings ChatGPT ads to 31 more countries

OpenAI began serving ads inside ChatGPT across 31 European markets on August 24, 2026, extending an advertising business that first launched in the United States about six months earlier. European advertisers will initially buy ad space through major advertising agencies, with self serve access planned for later.

OpenAI says roughly one in five ChatGPT queries already show direct commercial intent, meaning the person asking is close to making a purchase decision, with plenty more queries showing earlier stage interest that could still lead to a sale. The company has said user conversations stay private and are not shared directly with advertisers, and it has adjusted its privacy practices to meet European data protection rules.

Bringing ads into a chat assistant is still a new enough idea that most of the industry is watching closely to see how people react to seeing sponsored content mixed into an AI conversation, rather than in a separate search results page or social media feed. Marketers are already struggling to measure how much of their traffic and sales actually comes from AI chat tools, since a person might discover a brand in ChatGPT but complete their purchase somewhere else entirely, days or weeks later.

The average company now runs 13 AI agents, up from 5 last year

Salesforce's second annual Agentic Enterprise Index, based on usage data from its Agentforce platform between February 2025 and April 2026, found that the average number of active AI agents per organization jumped from 5 to 13 over that period, nearly tripling. The average time it takes a business to build and launch a new agent also fell by 53 percent.

An AI agent, in this context, is software that can complete a multi step task on its own, such as looking up a customer's order, checking inventory, and then issuing a refund, rather than simply answering a question in a chat window. Salesforce found that the average agent's skill set also roughly tripled over the year, growing from about 2 to 6 distinct actions it can take, letting a single agent handle work that once required stitching together several separate tools.

The report found a divide between industries: consumer facing sectors like retail favor large numbers of narrow, fast agents, with retail agent activity growing 18 times over during the analysis period, while more complex sectors such as manufacturing lean

toward fewer, more versatile agents that handle longer workflows. Because the data comes only from businesses already using Salesforce's agent platform, it reflects committed adopters rather than the average company, but it still shows a clear direction for where enterprise AI spending is heading.

Banks and payment apps start replacing chatbots with task completing agents

A roundup of AI agent developments this week highlighted banks and financial institutions moving beyond simple chatbots toward agents that can complete real tasks, citing work by Google and Singapore's DBS Bank on frameworks for controls, approval flows, and performance tracking around agent driven banking work.

A chatbot mostly answers questions, but a task completing agent can actually carry out a banking action, such as processing a transaction or updating an account, which raises the stakes considerably in a heavily regulated industry. That is why the emphasis in this shift is on audit trails, clear approval steps, and measurable performance, so a bank can prove exactly what an agent did and why, rather than just trusting it to get things right.

Large, heavily regulated institutions moving from experiments to production agent use is a signal that other vendors selling AI tools to banks and insurers will increasingly need to lead with governance and audit features, not just raw model capability, if they want to win contracts in finance. It also suggests that agentic AI is crossing from a novelty in retail and customer service into some of the most cautious, compliance heavy corners of the economy.

A small business payments agent goes fully live for every Cashfree customer

Indian payments company Cashfree moved its AI powered Super Agent, called Relay, from a limited beta that had been running since May 2026 to general availability for all Cashfree customers this week. Relay is built to automate payment operations for small and medium sized businesses.

Instead of a business owner manually checking on failed payments, reconciling transactions, or following up on overdue invoices, Relay is designed to handle that

routine payment administration work on its own, freeing up the owner's time for tasks that actually require a human decision. This fits a broader pattern of AI agents moving into the unglamorous back office tasks that small businesses often cannot afford to hire a dedicated person to handle.

Cashfree's move from a beta running since May to full availability for every customer suggests the company saw enough reliability and demand over those few months to commit to a wider rollout. It adds Cashfree to a growing list of financial technology companies building narrow, task specific agents around a single business function, payments in this case, rather than trying to build one general purpose assistant that does everything.

South Korea's Wrtn raises 72 million dollars for its AI entertainment app

South Korean AI company Wrtn Technologies closed a Series C funding round of about 72 million dollars, or 100 billion won, at a valuation of roughly 722 million dollars, bringing its total funding raised to about 166 million dollars. New investors Coreline Ventures and Eugene Asset Management joined existing backers Goodwater Capital, Antler, KDB, and Capstone.

Wrtn runs an AI entertainment platform aimed at North American audiences called OOC, which crossed 7.2 million dollars in monthly revenue within just three months of its May 2026 launch. The company expects its overall 2026 revenue to top 200 billion won, more than four times its 47.1 billion won in 2025, and it is preparing for a possible public listing in 2028.

Wrtn's rapid growth shows that AI focused entertainment and companionship style products, distinct from productivity tools like coding assistants or chatbots for work, are attracting serious investor interest and generating real revenue quickly once they find an audience. It is a reminder that not all of this year's AI funding activity is chasing frontier model labs or enterprise software, some of it is chasing consumer attention directly, and a smaller regional player can still put up striking growth numbers if it finds the right audience early.

Regulators and AI labs push for shared rules on what agents are allowed to do

A roundup of AI agent policy developments this week pointed to Google's Agent Payments Protocol, early work from the U.S. government's National Institute of Standards and Technology on agent identity and permissions, and a proposed federal bill called the AI AGENT Act as signs that industry and regulators are converging on rules for what an AI agent should be allowed to do on someone's behalf.

As agents start handling financial or legal actions, such as making a purchase or agreeing to a contract term, there needs to be a clear record proving that a specific action was actually authorized by the right person, with a time limit and a paper trail that can settle disputes without a long investigation. Right now, that kind of verification is inconsistent across different AI agent platforms, which creates real risk for any business letting an agent commit money or change account permissions.

This effort sits alongside the recent move by Google's A2A agent communication protocol to join Anthropic's Model Context Protocol under one shared governance structure, showing a broader push this year to standardize not just how agents talk to each other, but also how their actions get authorized and verified. For any business planning to let an agent handle real money or sensitive data, this area of authorization and audit trails is likely to matter more over time than which specific model powers the agent, since a fast, capable agent with no verifiable record of what it did is a much harder thing for a finance or legal team to sign off on than a slightly slower one that leaves a clean paper trail behind.

Quick Recap

Sony Music Publishing and Warner Chappell sued Anthropic over song lyrics allegedly used to train Claude.

OpenAI is cutting off Cursor's access to its models after SpaceX completed its acquisition of the company.

Claude Sonnet 5's discounted launch pricing ends today, reverting to standard rates.

ChatGPT is retiring its standalone DALL-E GPT, though ChatGPT Images continues.

California lawmakers are racing to pass roughly two dozen AI bills before adjourning tonight.

Amazon will shut down Mechanical Turk on September 30 after 21 years.

Marvell forecasts its AI data center chip business will hit 18 billion dollars by 2028.

OpenAI expanded ChatGPT advertising to 31 European markets.

Salesforce found the average company now runs 13 AI agents, up from 5 a year ago.

Banks are starting to move from chatbots to agents that can complete real tasks.

Cashfree's AI Super Agent, Relay, is now generally available to all customers.

South Korea's Wrtn Technologies raised 72 million dollars for its AI entertainment app.

Regulators and labs are pushing for shared standards on what AI agents can be authorized to do.

Frequently Asked Questions

What is the top AI news today?

The biggest story is Sony Music Publishing and Warner Chappell Music suing Anthropic over song lyrics they say were illegally used to train Claude, in a case that could carry billions of dollars in damages.

Why are Sony and Warner suing Anthropic?

The publishers allege Anthropic torrented, scraped, and downloaded tens of thousands of copyrighted musical compositions to train and operate its Claude models, and are seeking statutory damages of up to 150,000 dollars per infringed work.

Why is OpenAI cutting off Cursor?

OpenAI says it cannot be confident that SpaceX, which acquired Cursor for about 60 billion dollars, will follow its usage terms, citing Elon Musk's own admission that xAI had previously violated OpenAI's terms of service.

Is Claude Sonnet 5 getting more expensive?

Yes. Claude Sonnet 5's introductory pricing of 2 dollars and 10 dollars per million tokens ends on August 31, 2026, reverting to its standard rate of 3 dollars and 15 dollars per million tokens.

Is DALL-E being removed from ChatGPT?

The standalone DALL-E GPT persona inside ChatGPT is being retired on August 30, 2026, but the general ChatGPT Images feature that most people use to generate pictures is not affected.

Recommended Blogs

How to Use Claude AI

How to Use Google Gemini

ChatGPT Free for Beginners 2026

Best AI Tools for Coding 2026

What Is Agentic AI

Learn AI in 5 Minutes a Day

Keeping up with AI does not require reading every release note yourself. Unrot delivers the day's biggest AI developments in a 5 minute daily lesson, written in plain English for beginners, students, and working professionals who want to stay current without the jargon.