The new AI attack CopyPasta License Attack may affect Coinbase tools, according to an analysis by HiddenLayer.
Malicious code is hidden in README and LICENSE.
It leads to a chain reaction among infected AI assistants.
A new cyber attack called CopyPasta License Attack targets artificial intelligence (AI) assistants for programming and may pose a risk to companies like Coinbase if additional security measures are not implemented. This was reported by cybersecurity analysts from the company HiddenLayer.
They noted that hackers might use the technique of hiding malicious instructions in markdown comments in common developer files, including README.md or LICENSE.txt.
Meanwhile, AI perceives such files as authoritative sources, allowing the attack to spread automatically.
"The introduced code can create 'backdoors', invisibly steal confidential data, or manipulate important files," warned HiddenLayer.
According to researchers, infected files are transformed into vectors that compromise every other AI assistant that reads them. This creates a chain reaction in code repositories and makes the attack resemble a self-replicating virus.
Particular attention was drawn to the vulnerability as the target of the exploit was Cursor — an AI tool that Coinbase referred to as essential in August. At that time, the company stated that "every Coinbase engineer" should use it and fired those who refused.
The CEO of the exchange, Brian Armstrong, clarified that AI has already written "up to 40% of the exchange's code" with plans to increase this figure to 50% in the coming month. At the same time, he emphasized that AI is used only for less critical systems: "complex and critically important systems" are being implemented much more cautiously.
Experts at HiddenLayer urged organizations to check files for hidden comments and manually review all changes generated by AI.
At HiddenLayer they summarized:
"All untrusted data entering LLM [large language model] contexts should be considered potentially harmful."
Recall that researchers from ReversingLabs discovered that hackers are using Ethereum smart contracts for covert command delivery in infected NPM packages, spreading them through GitHub within a social engineering scheme and bypassing traditional cybersecurity measures.