Irony handed away quietly this week whereas Anthropic CEO Dario Amodei was in Australia lecturing native politicians on the significance of belief and security within the synthetic intelligence age.
The source code for Anthropic’s AI chatbot, Claude, leaked.
It was apparently “brought on by human error, not a safety breach”, the corporate stated, reassuring everybody they’ll really feel higher about belief and security with the tech giants in cost.
But it surely did sound a bit of like somebody walked out of Fort Knox, whereas telling everybody how secure their gold was, and left the entrance door open.
The code rapidly discovered its option to GitHub and the curious had been out of the blue trawling the more than 2,000 files containing 512,000 lines of code.
So builders did one thing very Anthropic – wolfed it up and commenced sharing it with others, usually in reworked variations.
Anthropic’s response was a “come to copyright” second. The corporate started issuing takedown notices to GitHub, utilizing the Clinton period Digital Millennium Copyright Act (DMCA).
“We issued a DMCA takedown in opposition to one repository internet hosting leaked Claude Code supply code and its forks,” the corporate stated.
All’s truthful use in love and AI
It was solely two years in the past that Amodei was utilizing the “truthful use” argument in a New York Times interview.
“I believe everybody agrees the fashions shouldn’t be verbatim outputting copyrighted content material. For issues which might be obtainable on the internet, for publicly obtainable, our place — and I believe there’s a powerful case for it — is that the coaching course of, once more, we don’t assume it’s simply hoovering up content material and spitting it out, or it shouldn’t be spitting it out,” he stated
“It’s actually way more like the method of how a human learns from experiences. And so, our place that that’s sufficiently transformative, and I believe the legislation will again this up, that that is truthful use.”
In that respect he was proper. In June 2025, a US court found in Anthropic’s favour, concluding whereas it was educated on copyrighted books to generative AI “quintessentially transformative” and qualifies as copyright “truthful use”.
However the choose additionally conclude that Anthropic intentionally selected to “steal” books slightly than search licensing agreements as a result of Amodei thought of doing issues the right approach a “authorized/apply/enterprise slog.”
His cofounder, Ben Mann, downloaded practically 200,000 books in 2021, figuring out they had been unauthorised copies – “that’s, pirated” the choose stated.
Three months later Anthropic agreed to settle a category motion involving greater than 500,000 authors for US$1.5 billion — the biggest copyright settlement in US historical past. That’s round $3000 an creator – and the full value is round 20% of Amodei’s estimated $7bn web price.
The ‘association’
His message to Canberra’s legislators about copyright was conciliatory. He shtick these days is a mixture of have your AI cake and eat it too, with few darkish clouds on the horizon amid the goodies and baddies – and guess which aspect Anthropic’s on.
“On copyright, particularly, I do know there was a sturdy debate on this nation, and we’re not right here to attempt to persuade you to vary your thoughts on this,” Amodei informed the politicians.
“I do assume rights holders do have authentic claims right here. We’re extra right here to speak about how can we arrive at an association that works for everybody and leaves everybody higher off.”
It’s price recalling that Tech Council of Australia chair Scott Farquhar stated final August in a dialogue about copyright that he was with people using Atlassian’s IP without paying for it if all of us get higher software program.
So certainly if I get the arms on Anthropic’s Claude code and rework it in a “quintessentially transformative” approach, I get do maintain the pet, don’t I Dario?
I’ll name him Honest Use. And even maintain the entrance gate shut so he doesn’t get out and begin humping the neighbour’s AI.
