X changes its terms to bar training of AI models using its content
Social network X, formerly known as Twitter, has updated its developer agreement to officially prohibit the use of its platform's public content for training artificial intelligence models. This move solidifies the platform's control over its vast dataset, particularly in light of its relationship with Elon Musk's own AI company, xAI.
The updated terms of service now include a specific restriction against this practice:
In an update on Wednesday, the company added a line under "Reverse Engineering and other Restrictions," a subsection of restrictions on use: "You shall not and you shall not attempt to (or allow others to) [...] use the X API or X Content to fine-tune or train a foundation or frontier model," it reads.
This policy change follows a series of adjustments and is seen as a strategic move to benefit its sister AI company:
This change comes after Elon Musk's AI company xAI acquired X in March — understandably, xAI wouldn't want to give its competitors free access to the social platform's data without a sale agreement. In 2023, X changed its privacy policy to use public data on its site to train AI models. Last October, it made further changes to allow third parties to train their models.
X is not alone in putting up walls around its data as the AI race heats up. Other technology companies have recently made similar changes to their policies to prevent unauthorized AI training:
Reddit has also put in place safeguards against AI crawlers, and last month, The Browser Company added a similar clause to its AI-focused browser Dia's terms of use.
As major platforms that host vast amounts of human-generated text and conversations increasingly restrict access for broad AI training, what might the long-term consequences be for AI development? Does this trend toward creating proprietary "data moats" risk stifling innovation and competition, potentially concentrating the future of advanced AI in the hands of a few powerful companies with exclusive data access?
(Score: 5, Funny) by aafcac on Friday June 06, @11:45PM (1 child)
We've had enough issues with AI going Nazi without the X posts.
(Score: 3, Troll) by Anonymous Coward on Saturday June 07, @12:46AM
Yes, but then there is xAI -- that one will still be able to regurgitate Elon's spew.
(Score: 1, Insightful) by Anonymous Coward on Saturday June 07, @02:14AM (4 children)
The only choice they have is to poison the data, put in control functions that blow up the "AI" servers, make the chatbot commit suicide
(Score: 1, Funny) by Anonymous Coward on Saturday June 07, @05:18AM (2 children)
Another obvious choice, sue the hell out of anyone that you catch crawling your site.
(Score: 2) by pkrasimirov on Saturday June 07, @07:11AM (1 child)
Good luck with China
(Score: 1, Informative) by Anonymous Coward on Saturday June 07, @01:41PM
Would anyone in China risk training an "AI" using data from X? Among other things, X/Twitter has been banned in China since 2009 -- see:
https://en.wikipedia.org/wiki/List_of_websites_blocked_in_mainland_China [wikipedia.org]
(Score: 5, Funny) by Thexalon on Saturday June 07, @12:39PM
If only AI crawlers were required to implement RFC 3514 [ietf.org] so they could be more easily identified.
"Think of how stupid the average person is. Then realize half of 'em are stupider than that." - George Carlin
(Score: 4, Troll) by jman on Saturday June 07, @11:01AM
(Score: 5, Insightful) by bzipitidoo on Saturday June 07, @05:25PM
It's pretty funny to see scrappy tech companies that got big by trampling upon their users' privacy and ignoring unenforceable rules turn all pearl clutching.
Once the data is out there, in public, there's no taking it back.