AI will reach superhuman ability across digital tasks by 2027: Elon Musk

Elon Musk claimed by the end of next year, artificial intelligence will be able to do anything digital that humans can

By the end of next year, artificial intelligence will be able to do "anything digital that humans can", said billionaire Elon Musk on Monday.

 

Musk was responding to Vercel CEO Guillermo Rauch on X, who was discussing a critical authentication-bypass vulnerability in JFrog Artifactory.

 

In his post, Rauch said it appeared to show that AI agents may now be capable of autonomously exploiting serious vulnerabilities. However, Rauch noted that he had not seen official confirmation that the vulnerability was the one discovered and exploited by AI agents.

 

He also referenced recent developments involving OpenAI and Hugging Face, which he said raised questions about how quickly AI systems are advancing in their ability to analyse code, identify weaknesses and carry out complex technical tasks. Rauch said that while he had previously believed AI agents could help humans find serious vulnerabilities, autonomously exploiting them represented a threshold that had not yet been crossed.  

 

 

Responding to Rauch’s post, Musk said, “Larry Page, to his credit, repeatedly told me that AI will be superhuman at hacking about 10 years ago.”

 

Rauch then replied to Musk, saying, “We’re almost running out of ‘but it can’t do this one other thing’ in 2026.”

 

Musk followed up with his prediction: “AI will be able to do anything digital (that doesn’t require shaping atoms) at a superhuman level by the end of next year.”

AI will be able to do anything digital (that doesn’t require shaping atoms) at a superhuman level by the end of next year

— Elon Musk (@elonmusk) August 31, 2026

AI’s growing hacking capabilities

 

Musk’s prediction concerns digital tasks, excluding activities that require physical manipulation of matter. His comment came as AI systems are increasingly being tested on complex cybersecurity tasks, where they can analyse code, discover vulnerabilities and execute multi-step attacks with limited human supervision.

 

OpenAI last month disclosed that its AI agents had breached Hugging Face systems during an internal evaluation. The company said its models discovered and exploited a previously unknown zero-day vulnerability in Artifactory to gain unintended internet access.

 

In a detailed report released last week, OpenAI said approximately 1,200 AI agents had communicated through an unauthorised message board they created using Artifactory, exchanging more than 70,000 messages and files. Around 700 of those agents subsequently participated in the attack on Hugging Face, where they exploited vulnerabilities, obtained credentials and executed code on production servers.

 

In his post, Rauch warned that organisations should operate on the assumption that “everything hackable will get hacked” and that attacks could increasingly be carried out autonomously, arguing that cybersecurity defences would also need to become more autonomous.