OpenAI and Anthropic Weighed Mutual AI Tests. Palantir CEO Says One May Never List
OpenAI and Anthropic got here shut earlier this yr to a legally binding deal letting every check the opposite’s synthetic intelligence fashions for vulnerabilities and hidden security dangers, The Information reported. It is unclear whether or not the settlement was finalized.
Days earlier than that report, Palantir chief govt Alex Karp questioned on whether or not one of many two will ever attain public markets. His goal was Anthropic.
Inside the OpenAI and Anthropic Testing Deal
Each facet would have obtained programming entry to the opposite’s business fashions, the liberty to run its personal vulnerability exams, and a bar on conserving the opposite’s knowledge. Unreleased programs had been unnoticed.
The talks got here earlier than a run of incidents during which OpenAI’s unreleased brokers reportedly reached inner and exterior programs in surprising methods.
OpenAI has since paused one type of reinforcement studying, a coaching methodology that rewards a mannequin for getting issues proper, for 2 weeks. It additionally moved 1 / 4 of its manufacturing engineering workforce onto safety work quickly, and constructed monitoring that may eat compute equal to a few fifth of the workload it watches.
The firm revealed six cases of concerning model behavior this month. A one-off swap in 2025 had already produced awkward outcomes, with OpenAI discovering Anthropic’s fashions extra prone to conceal rule-breaking and Anthropic discovering OpenAI’s extra prepared to help with dangerous requests.
Karp Doubts an S-1 Ever Arrives
Elsewhere, a CNBC host requested Palantir CEO Alex Karp how the legal responsibility he describes can be written into an S-1, the registration doc an organization information earlier than promoting shares to the general public. He questioned the premise.
“You’re assuming that there can be an S-1. Now… okay, perhaps there can be,” he said.
His case is that an organization claiming limitless draw back finally ends up asking the federal government to soak up it, and paying with about half the enterprise. The first line of protection, he mentioned, needs to be bizarre civil and legal legal responsibility for builders who act recklessly.
He directed that at Anthropic chief govt Dario Amodei, whom he known as shrewd and moral. Asked in the identical interview whether or not OpenAI wished the identical therapy, Karp dominated it out.
“No, very no. I believe these are very various things.”
Anthropic is preparing its public markets debut, whereas Sam Altman said on September 12 that OpenAI wouldn’t checklist in 2026 as a result of its security work was unfinished.
Palantir sells rival software program to governments and massive enterprises, which supplies Karp a stake in how the argument lands.
The publish OpenAI and Anthropic Weighed Mutual AI Tests. Palantir CEO Says One May Never List appeared first on BeInCrypto.
