Coinbase Blocks Cyberattack Targeting Open-Source AI Toolkit Agentkit
Coinbase thwarted a cyberattack targeting its open-source AI toolkit, agentkit, which aimed to inject harmful code into its CI/CD pipeline through GitHub permissions. Quick action from Coinbase and cybersecurity experts prevented significant damage.
- The incident was revealed on March 23 by Yu Jian of SlowMist, referencing a report from Palo Alto Networks' Unit 42.
- The attack occurred on March 14, 2025, involving the related toolkit onchainkit.
- The hacker exploited broad GitHub “write-all” permissions to insert dangerous code intended to collect internal data.
- Injected code lacked severe threats like remote control or reverse shell features but still posed risks.
- Coinbase's prompt response isolated the threat and applied protections, averting further issues.
- This incident highlights the importance of securing open-source tools in the growing crypto sector.
- Over $1.5 billion in crypto losses from hacks and exploits have been reported this year, according to DeFiLlama.
- Yu Jian advised developers using tools like reviewdog or tj-actions to conduct thorough system reviews.
The attack underscores potential vulnerabilities within the crypto industry, where Coinbase plays a critical role as a leading exchange and custodian for Bitcoin ETFs.