In short: ChatGPT maker OpenAI is being investigated by the Federal Commerce Fee to find out if the corporate’s chatbot has damaged shopper legislation by damaging reputations and placing private knowledge in danger. The FTC additionally desires data on two safety incidents from earlier this 12 months.
The company despatched a 20-page demand letter to OpenAI for data about the way it addresses dangers associated to its AI fashions, writes The Washington Post. The Civil Investigative Demand additionally requests particulars in regards to the firm, its AI mannequin and coaching, the way it offers with privateness and immediate injection assaults, API and plugin integrations, and knowledge assortment.
One of many statements particularly asks what OpenAI has achieved to deal with the hazard of its merchandise producing false, deceptive, or disparaging details about actual folks. Generative AIs are infamous for his or her tendency to hallucinate, in fact.
The Reg points out that OpenAI is not any stranger to defamation lawsuits. A Georgia radio host sued the corporate over allegations that ChatGPT falsely related his title with a felony subject. And in April, an Australian mayor threatened authorized motion claiming ChatGPT accused him of involvement in a international bribery scandal. The chatbot additionally falsely accused a US legislation professor of sexual assault based mostly on a fabricated Washington Publish article.
OpenAI hasn’t been very open in disclosing what knowledge it makes use of to coach its massive language fashions powering ChatGPT. The FTC desires to know if it has been scraped from the web or if it has been purchased from third events. The company additionally asks which web sites the info has been taken from and what was achieved to stop private data from getting used as coaching knowledge.
The info-scraping practices of generative AI corporations has been a controversial subject lately. Elon Musk threatened to sue Microsoft, which has invested billions into OpenAI, for illegally utilizing Twitter knowledge – it was additionally the claimed motive behind limiting customers’ tweet views. Reddit, in the meantime, noticed a mass person revolt after it began charging for entry to its API in response to knowledge scraping by AI corporations.
The FTC additionally desires data and data associated to safety incidents earlier this 12 months. One was a bug that allowed some customers to see payment-related data, the opposite uncovered the titles of users’ chat histories.
we constructed GPT-4 on prime of years of security analysis and spent 6+ months after we completed preliminary coaching making it safer and extra aligned earlier than releasing it. we defend person privateness and design our methods to be taught in regards to the world, not personal people.
– Sam Altman (@sama) July 13, 2023
OpenAI CEO Sam Altman stated the corporate will work with FTC. He added that it protects person privateness and designs its methods to be taught in regards to the world, not personal people.
The FTC has already handed massive fines to Meta, Amazon, and Twitter over allegations they violated shopper safety legal guidelines. OpenAI may face an analogous punishment whether it is discovered to have achieved the identical, a transfer that would go away different generative AI corporations sweating.
Earlier his month, Google updated its privateness coverage to explicitly state that the corporate reserves the suitable to gather and analyze just about something folks share on the internet to coach its AI methods.
