AI lawsuits and bank agent breaches sharpen the case for rules
- Reservoir Media filed federal lawsuits against Anthropic and Suno on 22 October 2026 over unauthorised training on copyrighted works.
- Anthropic has said it disputes the claims brought in the filings.
- Major US banks reportedly detected unauthorised third-party AI agents accessing customer accounts.
- An ABC News/Washington Post/Ipsos poll found 57 per cent of Americans consider AI bad for society overall.
Reservoir Media filed federal lawsuits against Anthropic and Suno on 22 October 2026, as Fox News’s roundup of that day’s AI developments records, alleging that both companies trained models on copyrighted works without permission. Madonna, Johnny Cash and Joni Mitchell are among the catalogues named. Anthropic says it disputes the claims. The names matter less than what the filings set in motion: the discovery process. Training corpora, licensing correspondence and internal decisions about what counted as permission will all be examined under oath. And the answers will not stay confined to two defendants.
The mechanics
The suits are civil actions. Nothing changes for users on the day of filing. What changes is the paper trail. Companies that scrape or license content now have a live reason to keep records that survive scrutiny. The absence of such records becomes an evidentiary problem, not a rhetorical one.
Money and security moved in the same roundup. Ron DeSantis, Florida’s governor, warned that rogue AI agents could drain bank accounts. He cited reporting that major US banks had detected unauthorised third-party agents accessing customer accounts, and argued that criminal liability for executives might deter the conduct. He also conceded that punishment after the fact is inadequate. Arista Networks warned of active exploitation of CVE-2026-114625 in on-premises VeloCloud Orchestrator. Microsoft attributed a supply chain campaign against thousands of ConnectWise tenants to a Chinese state-affiliated group using an open-source agentic framework. An ABC News/Washington Post/Ipsos poll put public disapproval of AI at 57 per cent, with partisan disagreement about personal benefit.
These items describe one problem from several angles. Autonomous systems act at machine speed while oversight stays human, retrospective and slow. In each case, enforcement depends on logs that the operator controls and may not keep.
For education and research, the practical lesson is documentary rather than doctrinal. A model trained on properly licensed material can show its sources. One trained on an unexplained corpus cannot. Institutions adopting AI tools for teaching or assessment should expect the questions the courts are asking, and should be able to answer with a licence, a date and an audit trail rather than an assurance. Students working with generated text face the same test in miniature. Provenance is the difference between a defensible draft and an unverifiable one.
What to watch
Three signals will show whether any of this hardens into rules: whether the Reservoir claims survive early motions to dismiss, whether congressional committees produce legislative text rather than further hearings, and whether banks publish incident counts that can be checked. Until one of those lands, the public record consists of allegations, vendor advisories and a poll. None of it settles who is accountable when an agent acts on its own.