AI Coding Tool Used by Coinbase Exposes Firms to Self-Spreading Malware

A newly disclosed vulnerability in an AI-powered coding tool favored by Coinbase has raised alarms across the cybersecurity and crypto communities.

Key Takeaways:

• A new AI coding exploit can silently spread malware across entire codebases using hidden markdown instructions.
• Coinbase’s preferred tool, Cursor, is among several AI assistants shown to be vulnerable.
• CEO Brian Armstrong’s aggressive AI rollout has sparked backlash from developers and security experts.

According to cybersecurity firm HiddenLayer, the flaw allows attackers to stealthily inject malicious code that can spread across an entire organization’s codebase with minimal user interaction.

The attack, dubbed the “CopyPasta License Attack,” exploits how AI tools interpret common developer files like LICENSE.txt and README.md.

By embedding harmful instructions in markdown comments, often hidden from rendered views, attackers can manipulate AI code assistants into propagating malware without developers realizing.

“Injected code could stage a backdoor, exfiltrate sensitive data, or manipulate critical systems, all while remaining buried deep inside files,” HiddenLayer said in a Thursday report.

The firm demonstrated the exploit using Cursor, the AI coding assistant reportedly adopted by every Coinbase engineer as of February.

HiddenLayer said similar vulnerabilities were present in other tools including Windsurf, Kiro, and Aider.

The concern comes just a day after Coinbase CEO Brian Armstrong claimed that AI now writes up to 40% of the company’s code, a figure he aims to push to 50% next month.

The announcement drew criticism from cybersecurity experts, developers, and crypto insiders who warned of the risks tied to mandated AI adoption.

“This is a giant red flag for any security-sensitive business,” said Larry Lyu, founder of decentralized exchange Dango.

Carnegie Mellon professor Jonathan Aldrich called the policy “insane,” adding that he would not trust Coinbase with his funds after hearing it.

~40% of daily code written at Coinbase is AI-generated. I want to get it to >50% by October.

Obviously it needs to be reviewed and understood, and not all areas of the business can use AI-generated code. But we should be using it responsibly as much as we possibly can. pic.twitter.com/Nmnsdxgosp

Delphi Consulting’s Ashwath Balakrishnan called the push “performative and vague,” while Bitcoiner Alex Pilař stressed that Coinbase, as a major crypto custodian, should prioritize security over AI adoption metrics.

Armstrong has defended the move, saying AI-generated code must still be reviewed and is not used in all parts of the business.

In a blog post, Coinbase’s engineering team clarified that AI use is more common in front-end and less-sensitive systems, while “system-critical exchange systems” remain more cautiously managed.

However, Armstrong admitted during a podcast with Stripe co-founder John Collison that he had enforced AI onboarding at Coinbase, going as far as firing engineers who refused to use the tools.

“I went rogue,” Armstrong said. “They got fired.”

As reported, TIME has recognized Coinbase as one of 2025’s 100 Most Influential Companies, labeling the crypto exchange a “disruptor” for its significant role in shaping US digital asset policies and markets.

TIME noted the exchange as a key driver behind the industry’s policy efforts and predicted Coinbase could become the central hub for crypto trading in the US.

Beyond the US, Coinbase is broadening its reach in Europe, securing a license under the EU’s MiCA regulatory framework through Luxembourg’s financial regulator.

The post AI Coding Tool Used by Coinbase Exposes Firms to Self-Spreading Malware appeared first on Cryptonews.