A Musk-Telegram AI pact sets the stage for more drama

By Parmy Olson

It’s hard to find a pair of tech billionaires more alike than Elon Musk and Pavel Durov. Both are staunch libertarians who run large social media platforms with minimal content moderation, and both champion themselves as defenders of free speech. Both are also pro-natalists, with Durov claiming to have fathered more than 100 biological children through sperm donation.

It’s an ideological match made in heaven — and perhaps in business too. Durov this week announced a one-year partnership with Musk’s xAI, in which the latter artificial intelligence lab would pay Telegram $300 million in cash and equity to integrate its Grok chatbot into the messaging app. In essence, Telegram gets whizzy new AI features while Grok gets distribution on a platform with roughly 1 billion users, double the number of people on X. But there are a couple of snags.

Durov, for a start, appears to believe an oral contract is as good as a written one. Hours after Durov announced the partnership and its financial terms on X, Musk witheringly replied: “No deal has been signed.”

Musk may not be in the best of moods. With the true impact of his Department of Government Efficiency still unclear, he left his post at the White House this week. Now he’s turning back to a business empire in flux, having leaned on AI to balance his books. The generative AI boom has pushed the value of two-year-old xAI to $80 billion, and Musk recently merged that company and X (formerly known as Twitter and worth far less) under a single entity, xAI Holdings.

That gives X some much-needed stability, but xAI still needs to make money, and the Durov deal could help. More users of Grok mean more opportunities to sell subscriptions for the AI tool, which cost between $3 and $40 a month. And there’s a historical precedent for this kind of distribution partnership to work. Dell Inc. made a killing from pre-installing Microsoft Corp.’s Windows on its PCs, while Apple Inc. has earned $20 billion a year from Alphabet Inc. by making Google the default search tool on iPhones.

There’s another tantalizing possibility for Musk: scraping messages and content across Telegram and using both to train Grok to be even smarter. A key feature of building today’s generative AI systems is that priming them with more data leads to better capabilities, and until now Grok has benefited from training on the mass of text (and bilge) on X.

But Durov seems to have nixed that possibility. When a user on X asked Durov if he’d allow such scraping, he replied that user privacy was “paramount” and that Grok would only extract messages people shared with the bot directly. Perhaps this was Durov’s way of preempting Musk from seeking access to Telegram data — hence going public on the partnership so early. And maybe we shouldn’t be surprised if the two hash out the issue out publicly.

That isn’t the only type of drama to expect from this pairing. Should Grok get plugged into Telegram’s app this summer, it will be one of the deepest known integrations of an AI tool into a major Western mobile messaging and broadcasting service. You can’t use AI to edit tweets on X or messages in WhatsApp and Facebook Messenger, even though all those platforms now have an AI assistant of some sort. But on Telegram, you’ll be able to do just that.

A video demo posted by Durov on Wednesday shows a Telegram user typing the word “great” in a text to their work colleagues. Grok offers options to “improve,” “expand” or “change tone.” When the user taps “expand,” their message turns into: “Excellent, team effort at its best. Keep up the good work!”

This at first simply looks like another example of AI bringing blandness and vapidity to human conversation. Apple and Google both offer suggested AI replies in their email software. But using that functionality to edit mobile text messages in real time pushes the technology further into the mechanics behind human voice and intention.

It also portends a darker impact on a platform like Telegram, where content rules are scant. The app is well-known for hosting extremist content and conspiracy theories, with neo-Nazi groups in the US using it to organize rallies, while some of last year’s anti-immigration riots in the UK were promoted on the platform.

Far-right groups, meanwhile, have found ways to “jailbreak” AI bots like Grok, getting them to generate content that breaks the rules of other social media firms; for instance, generating a photorealistic image of mountains that spell out “The Jews did 9/11,” according to the International Centre for Counter-Terrorism, which singled out Grok as being relatively easy to exploit. Musk has yet to publicly respond to criticisms about jailbreaking Grok.

If Grok allows Telegram to create more persuasive memes and other forms of propaganda at scale, that could make it an even more powerful tool for spreading toxicity, from disinformation to hate speech to other odious content.

Musk and Durov have kicked off a mini corporate drama of their own, but the real chaos could unfold on the platform itself, and it won’t be pretty.