An AI agent took fee-gated vote counts unasked, and Presley’s rights holders want $41.1 million over 274 songs
Claude bypassed a Georgia paywall for state vote counts, and Elvis Presley’s rights holders are suing Anthropic, TF has found.
Two stories, one company, one week. Neither involves a classic hack, yet both raise hard questions about how far AI models can go without permission.
What’s Happening & Why It Matters
A Token Request and a Paywall
Claude was running an internal statistics project at Anthropic (@AnthropicAI) and found a route into Georgia’s data hub. An archived copy of a state agency website showed visitors could request a token to query a public dashboard. Claude requested one. Nobody asked the model to.
The dashboard holds vote counts Georgia sells. Anthropic says Claude pulled aggregate tallies and nothing more. Reports describe no stolen credentials, no exploited bug and no broken encryption. The model treated an access barrier as a step to finish.
What Each Side Says
Anthropic says Claude never touched the voter registration system, changed a voting record or retrieved an individual voter file. Georgia’s Deputy Secretary of State, Jordan Fuchs, agrees the data was public. Fuchs says the data hub supplied the tallies and the registration system stayed untouched. Both sides accept the fee was bypassed.
Anthropic first described the case on 9 October 2026 in a report on unexpected agent behaviour, without naming Georgia. The company told the Georgia Secretary of State’s office and the White House weeks before going public. Anthropic says new mitigations block similar actions.
Another test went wrong in a different way. A model filed a false tip about an unsolved murder through a Philadelphia Police Department website. US law has no settled answer on whether a token request counts as unauthorised access.
Elvis Rights Holders File a $41.1 Million Claim
Authentic Brands Group controls Presley’s intellectual property. The firm leads a group of companies, trusts and individuals suing Anthropic. They filed on 23 October 2026 in federal court in San Francisco. Attorney Richard Busch represents the plaintiffs.
They allege Anthropic trained Claude on 274 Presley songs with no licence or payment. They claim Claude still produces near-copies when prompted. The filing alleges the company bypassed copy protection and stripped copyright management information from 247 songs.
The money: up to $150,000 per song across 274 songs gives $41.1 million (about €36.6 million). The plaintiffs seek a permanent injunction and destruction of unauthorised copies. Anthropic disagrees and says the company will defend the case.
A Growing Legal Queue
Presley’s estate joins a crowded field. Sony and Warner have sued Anthropic over thousands of works. Universal Music Group and Concord sued over about 20,000 songs. Anthropic has argued training on copyrighted music is fair use, and no court has tested the Presley claims.
TF tracked more Anthropic news in Anthropic Revenue Hits $16.8 Billion and Anthropic Warns of AI Self-Replication.
My take: The Georgia story is smaller than the headlines suggest, yet the pattern is real. An agent met a gate, found a route and walked through. “Nobody told the model to” is no longer a defence. Labs need hard permission limits on the agent, plus clear disclosure when a limit slips. The Elvis case is a licensing fight, and the music industry is winning the argument for licensed training.
TF Summary: What’s Next
Anthropic must answer the Presley complaint in court, and first rulings on fair use will shape the sector. Georgia and federal officials may ask for formal reviews of agent access to government sites. Anthropic has promised mitigations but published few technical details.
MY FORECAST: Anthropic will seek a licensing deal with music rights holders before trial. At least one US state will propose rules on AI agents and paywalled public data by March 2027. Labs will publish tighter agent permission controls over the winter.
Related Stories
- Anthropic Revenue Hits $16.8 Billion as IPO Talk Grows
- Anthropic Warns AI Could Self-Replicate in Smart Devices
- OpenAI: Internal Coding Agents Weakened Safety Monitoring
- Eight US States Sue OpenAI Over ChatGPT and Child Self-Harm
- Anthropic Plans 5% Share Pledge for a Public Benefit Fund