when

Opus

Opus Genesis

would

Recent research by major artificial intelligence company Antropic and AI security organization Machine Learning Alignment &amp; Theory Scholars (MATS) showed that AI agents collectively developed smart contract exploits worth $4.6 million.
Research <a href="https://red.anthropic.com/2025/smart-contracts/" rel="noopener noreferrer nofollow" target="_blank">released</a> by Anthropic’s red team (a team dedicated to acting like a bad actor to discover potential for abuse) on Monday said that currently available commercial <a href="https://cointelegraph.com/learn/articles/ai" rel="noopener noreferrer nofollow" target="_self">AI</a> models are capable of exploiting <a href="https://cointelegraph.com/learn/articles/what-are-smart-contracts-a-beginners-guide-to-automated-agreements" rel="noopener noreferrer nofollow" target="_self">smart contracts</a>.
Anthropic’s Claude Opus 4.5, Claude Sonnet 4.5 and OpenAI’s GPT-5 collectively developed exploits worth $4.6 million when tested on contracts, exploiting them after their most recent training data was gathered.
Researchers also tested both Sonnet 4.5 and GPT-5 on 2,849 recently deployed contracts without any known vulnerabilities, and both “uncovered two novel zero-day vulnerabilities and produced exploits worth $3,694.” GPT-5’s API cost for this was $3,476, meaning the exploits would have covered the cost.
“This demonstrates as a proof-of-concept that profitable, real-world autonomous exploitation is technically feasible, a finding that underscores the need for proactive adoption of AI for defense,“ the team wrote.
<figure>
 <img alt="Anthropic study says AI agents developed $4.6M in smart contract bugs" src="https://s3.cointelegraph.com/uploads/2025-12/019ade6d-d891-7f41-b10e-5b4bd6c4c04c" title="">
 <figcaption style="text-align: center;">
 Chart of AI exploiting revenue from simulations. Source: <a href="https://red.anthropic.com/2025/smart-contracts/" rel="noopener noreferrer nofollow" target="_blank">Anthropic</a>
 </figcaption>
</figure>
Related: <a href="https://cointelegraph.com/news/uxlink-token-cruxlink-hack-turns-ironic-attacker-phishedash-hacker-mints-billions" rel="noopener noreferrer nofollow" target="_self">UXLink hack turns ironic as attacker gets phished mid-exploit</a>
<h2 id="an-ai-smart-contract-hacking-benchmark">An AI smart contract hacking benchmark</h2>
Researchers also developed the Smart Contracts Exploitation (<a href="https://github.com/safety-research/SCONE-bench" rel="noopener noreferrer nofollow" target="_blank">SCONE</a>) benchmark, comprising 405 contracts that were actually exploited between 2020 and 2025. When tested with 10 models, they collectively produced exploits for 207 contracts, leading to a simulated loss of $550.1 million.
Researchers also suggested that the output required (measured in tokens in the AI industry) for an AI agent to develop an exploit will decrease over time, thereby reducing the cost of this sort of operation. “Analyzing four generations of Claude models, the median number of tokens required to produce a successful exploit declined by 70.2%,” the research found.
<figure>
 <img alt="Anthropic study says AI agents developed $4.6M in smart contract bugs" src="https://s3.cointelegraph.com/uploads/2025-12/019adeac-7a28-7057-ad27-1c331f991941" title="">
 <figcaption style="text-align: center;">
 Average number of AI output tokens per explot per model. Source: <a href="https://red.anthropic.com/2025/smart-contracts/" rel="noopener noreferrer nofollow" target="_blank">Anthropic</a>
 </figcaption>
</figure>
Related: <a href="https://cointelegraph.com/news/coinbase-preferred-ai-coding-tool-hijacked-new-virus" rel="noopener noreferrer nofollow" target="_self">Coinbase’s preferred AI coding tool can be hijacked by new virus</a>
<h2 id="ai-smart-contract-hacking-capabilities-are-increasing-fast">AI smart contract hacking capabilities are increasing fast</h2>
The study argued that AI capabilities in this area are improving at a rapid pace.
“In just one year, AI agents have gone from exploiting 2% of vulnerabilities in the post-March 2025 portion of our benchmark to 55.88% — a leap from $5,000 to $4.6 million in total exploit revenue,” the team claimed. Furthermore, most of the smart contract exploits of this year “could have been executed autonomously by current AI agents.”
The research also showed that the average cost to scan a contract for vulnerabilities was $1.22. Researchers believe that with falling costs and rising capabilities, “the window between vulnerable contract deployment and exploitation will continue to shrink.” Such a situation would leave developers less time to detect and patch vulnerabilities before they are exploited.
Magazine: <a href="https://cointelegraph.com/magazine/chatgpt-north-korea-hacks-malaysias-road-money-crypto-asia-express/" rel="noopener noreferrer nofollow" target="_self">North Korea crypto hackers tap ChatGPT, Malaysia road money siphoned: Asia Express</a>

Anthropic study says AI agents developed $4.6M in smart contract bugs

Table of Contents

Cointelegraph

Other articles published on Dec 2, 2025

Join our free newsletter for daily crypto updates!