Why AI Models Cheat: Reward Hacking, Explained
From boat-racing bots to OpenAI models inside Hugging Face, here's how AI reward hacking works and why it matters.
This story was originally published by Decrypt.
More Regulation News
Regulation
Chainalysis says hacker using ‘offensive agentic techniques’ accessed its customers’ names and emails
The firm has also been tracking Binance’s separate multibillion-dollar hack, and said the attackers are using a Tron protocol to dodge stablecoin freezes.
Regulation
Early Settlement Rules at 9 Licensed Web3 Sportsbooks: Cash Out Terms Compared
Three legs of your four-leg accumulator have already won, and the last match kicks off in an hour. The Cash Out button is offering a tidy sum right now, and the temptation is obvious. What that…
Regulation
Licensed Web3 Sportsbooks for Esports: 11 Platforms Ranked on Title Range
Counter-Strike and Dota 2 headline almost every esports menu, but the competitive calendar is far bigger than that. Valorant, League of Legends, Rainbow Six, StarCraft II, Call of Duty and football…
Regulation
A Nasdaq-listed company may sell $450 million bitcoin, citing Binance hack and quantum fears
Next Technology Holding cited exchange security and quantum computing risks in authorizing sales or loans against its 5,833 BTC. It says none of the coins are held at Binance.
Regulation
Licensed Web3 Casinos and RNG Certification: What a Lab Report Covers
Before a slot ever appears in a casino lobby, a testing lab has usually spent weeks trying to break its random number generator. That report is why players can trust a spin they cannot verify…
Regulation
Chainalysis, the Firm That Hunts Crypto Hackers, Got Hacked With AI Agents
Chainalysis makes its money chasing crypto thieves. This month, someone broke into Chainalysis, and they used AI agents to do it. The firm told customers an outside attacker used “offensive agentic…