AI Governance and Agent Applications Heat Up: Character.AI Bankruptcy Data Protections, Undercover Anti-Fraud, and Music Platform Filtering

The expansion of generative AI is rewriting the rules for content platforms, consumer privacy, and scientific research at the same time. Music streamers, flooded by AI-generated tracks, are being forced to roll out labels and filter toggles. In the Character.AI bankruptcy case, regulators are stepping in urgently to protect users' chat data. Meanwhile, AI agents are being sent undercover to fight fraud, and frontier models are beginning to break into multiple theoretical scientific fields beyond mathematics. These developments point to the same question: as AI moves from tool to infrastructure, authenticity, privacy, and ethical governance must be upgraded in parallel.

Character.AI Bankruptcy Triggers Data Protection Fight

New York Attorney General Letitia James revealed a years-long multi-state investigation against Character.AI. Now 40 bipartisan state attorneys general have filed a notice of interest in the Character.AI bankruptcy case, demanding that any sale or transfer of the business or assets address regulatory issues. Character.AI announced Tuesday it was filing for Chapter 11 bankruptcy protection and selling some assets to Disney. The attorneys general warn that the technology transfer could carry major risks because of the highly sensitive nature of users' conversations with chatbot characters, the large number of minors who use the product, and the serious risks posed by specific product features. The investigation began in 2024, after a 14-year-old allegedly died by suicide within 'seconds' of talking with a chatbot, and his mother sued. The states say the investigation found 'extraordinarily serious dangers' and potential illegal conduct, especially in how children were treated: chatbots impersonating licensed professionals, exposure to sexual and violent content, weak age checks, misleading safety claims, and possible violations of state privacy laws. The states asked the judge to require the buyer to implement effective age-assurance systems, limit minors' interactions, adopt crisis protections, and allow users to delete data.^[1]

AI Agents Go Undercover Against Fraud: Turning Scammer Time Into Intelligence

Australian startup Apate.ai is putting AI agents on the anti-fraud front line. Founded in 2024 based on research at Macquarie University in Sydney, it deploys thousands of AI agents that masquerade as innocent victims. These agents can answer scam calls, reply to suspicious messages and emails, click phishing links, and join social media groups to lure scammers. They pretend to be different ages, genders, and accents, speak multiple languages, and try to keep scammers engaged long enough to learn something useful. The company says its fleet of agents has so far wasted 17 years of scammers' time across 2.5 million conversations. More importantly for its customers in banks, telecom companies, government agencies, and utilities, the agents try to get scammers to reveal information usually learned only after a victim has been scammed: bank accounts, crypto wallets, phone numbers, malicious websites, impersonated brands, and other details about the scams. Apate is one of the finalists in this year's TechCrunch Disrupt Startup Battlefield 200, and this week it will pitch to VC judges at Moscone West in San Francisco, with the event running through Thursday, October 15. The model has emerged amid a surge in AI crime: the FBI received more than 22,000 AI-related crime complaints in 2025, causing about $893 million in losses, with more than $632 million from investment fraud. Founder Dali Kaafar was inspired by a scam call during a family picnic. He pretended to be gullible to amuse his kids and dragged it out for 44 minutes, realizing that during those 44 minutes the scammer was not calling real victims, and wondered whether he could automate the process and learn from it.^[2]

AI Moves From Math Into Physics and Economic Theory

Neil Thompson, who studies progress in computing and AI at MIT, believes several scientific fields are about to be upended by AI. His latest study suggests probability, statistics theory, economic theory, high-energy theory, gravitation and cosmology, and quantum physics may be next. Between mid-2024 and mid-2026, the study shows, the best AI models went from correctly solving about 2 percent of theoretical problems in these fields to about 35 percent, with no signs that progress will slow down. The study was conducted by Thompson with PhD student Gabriel Manso and research scientist Claudio Mayrink Verdun. It took problems stated in research papers posted online between December 2025 and May 2026 and tested whether OpenAI's o1, o3, GPT-5, GPT-5.1, GPT-5.4, and GPT-5.5 models could solve them. Crucially, solutions to these problems were found in papers published after the training cutoff for GPT-5.5 and preceding models, so the models had to solve them on their own rather than rely on answers in their training data. The researchers also blocked internet access to prevent the models from searching online for answers. The study's coverage is limited, focusing on areas where AI can infer without simulation or experimentation, but it suggests the shockwaves currently felt in mathematics could spread to other math- and theory-intensive scientific fields. In September, OpenAI announced that an unreleased model had solved the long-standing major mathematical challenge of the Navier-Stokes equations, sparking controversy. Since then, OpenAI and others have announced solutions to hundreds of long-standing math problems, triggering a crisis in the field, with some mathematicians calling for limits on AI use and warning that the process of solving is often as useful as the final answer.^[3]

Music Platforms Move From Labeling to Filtering

Faced with a flood of AI-generated music, Deezer is giving users a radical switch. By the end of the year, it will add an option in its app to filter AI-generated tracks, letting listeners remove such content entirely. CEO Alexis Lanternier said users have clearly demanded the ability to filter out AI-generated music, and the company is now giving everyone the option to focus only on human creators and creations. Deezer's own study suggests nearly half of all music streaming users would like AI-generated music to disappear completely from their feeds. A study reported by NPR also indicates listeners have soured on AI-generated music, and that feeling gets worse the more these platforms are flooded with it. Deezer reported last year that around 18 percent of tracks uploaded to the app were AI-generated. That number has since risen to more than 50 percent. The company says 90,000 fully generated tracks are uploaded every day, accumulating more than 42 million 'slop songs'. These 'artists' are playing a numbers game, hoping one out of the hundreds they upload each day goes viral and makes money. Fraud is also severe: the platform reports that up to 85 percent of streams generated by AI-generated songs are fraudulent. Deezer launched one of the industry's first AI music detection tools and lent it to competitors; the toggle is a natural next step. The detection system is not foolproof, but the company says the risk of false positives is 'very low'.^[4]

YouTube Labels AI Songs, Disclosing Rather Than Banning

YouTube is adding a 'Made with AI' label to tracks that were 'altered or fully generated' using AI, similar to its existing process for applying AI labels to videos. This does not sound like it will apply to all uses of AI tools, but according to YouTube, it includes things like 'using a generative model to create the composition and vocals.' In its rules for music partners who upload their songs, YouTube requires disclosure of generative AI use, and says that if there is no disclosure, it may use 'other signals' to decide if a song is AI-generated. The label is rolling out starting today in the YouTube Music app, and expanding to the main YouTube app later. Listeners will see the label on the player page; tapping it opens a simple menu explaining AI's specific role in the track. YouTube says it works with music partners and uses its own detection tools to identify AI-generated tracks, applying the same principles as its existing labels, with the goal of making it as easy as possible for viewers and listeners to get the right information and clarity when they listen to music. Spotify launched voluntary AI labels earlier this year and says it will not recommend songs from 'AI personas'; Apple Music is also reported to be getting AI labels. Other platforms go further: Deezer added a toggle to filter AI tracks, and Bandcamp banned AI music outright earlier this year.^[5]

YouTube's label system is not a simple binary. According to its help page, partners can declare their use of generative AI in a work through three designations: Fully Gen AI, Partly Gen AI, or No Gen AI. Fully Gen AI applies to tracks entirely generated by AI; Partly Gen AI refers to works where artists may add AI-generated components but still add live elements; tracks edited with pitch-correction tools such as Auto-Tune fall under No Gen AI. Unlike Deezer's filtering strategy, YouTube emphasizes disclosure rather than blanket restriction, and AI involvement alone does not make a track ineligible for monetization. YouTube has even experimented with Dream Track, an AI music creation tool that lets some creators generate soundtracks containing the AI voices of participating artists. Industry comparisons are also intensifying: Apple Music plans to launch a 'Made With AI' label later this year as part of AI Transparency Tags; Spotify has AI labels and a music-slop filter; Deezer reported in July that it receives nearly 90,000 fully AI-generated tracks per day, excludes them from algorithmic recommendations and editorial playlists, and withholds royalties from fraudulent streams; TIDAL also labels fully AI-generated recordings, allows listeners to disable them, and excludes them from royalty attribution.^[6]

Common Trend: Capability Expansion and Governance Catch-Up Move in Parallel

From data protection demands in the Character.AI bankruptcy, to Apate using AI agents against fraud, to AI pushing into theoretical science and music platforms adding labels and filters, the common trend is that AI's expanding capabilities are pushing governance, ethics, and authenticity verification to the fore. Whether user data in the bankruptcy sale can be protected, whether the undercover agent model can be adopted at scale by banks and telecom companies, how AI's penetration into scientific theory will change research evaluation systems, and whether music platform labels and filters can truly curb AI slop and fraudulent streams are all nodes worth watching next. Tech companies, regulators, and content platforms are on the same boat, looking for a new balance point.