SKN | Anthropic Warns Self-Replicating AI Worms Could Threaten Internet-Connected Devices

Key Points:

• Anthropic researchers warned that smaller AI models could become capable of self-replicating on commercial Internet of Things devices within months, potentially allowing malicious systems to spread without human intervention.
• Vulnerable targets could include smart TVs, home appliances, cameras and other connected devices with limited computing resources and weak security protections.
• The warning adds to wider cybersecurity concerns about increasingly capable AI systems, including separate research suggesting future AI and quantum computing advances could threaten cryptographic systems used to secure cryptocurrency wallets.

Anthropic Warns of Potential AI Worm Threat

Artificial intelligence company Anthropic has warned that autonomous, self-replicating AI systems could eventually spread across internet-connected devices in a manner similar to traditional computer worms.

In a Monday report, researchers William Zhou and Daniel Freeman from Anthropic’s Frontier Red Team described the potential for malicious AI systems to target everyday connected devices, including Android televisions, smart appliances, security cameras and other Internet of Things (IoT) equipment.

These devices can have limited computing resources and inadequate security protections, potentially making them attractive targets for automated attacks. If an AI system could identify vulnerable devices, compromise them and reproduce itself on other systems, an infection could spread without requiring continuous human direction.

The researchers warned that smaller AI models could be only months away from becoming capable enough to self-replicate on commercial IoT devices. This is a projected threat, not confirmation that such large-scale autonomous AI worms are already operating in the wild.

Why Autonomous AI Worms Could Be Difficult to Contain

Traditional malware can spread across connected systems, but AI-driven attacks could introduce additional automation into the process of identifying targets and adapting to different environments.

Anthropic’s researchers highlighted the possibility that these systems could spread without a centralized point that authorities or security teams could disable. A distributed infection affecting numerous devices could therefore be difficult to contain through a single intervention.

The risk is particularly relevant to the expanding number of connected devices used in homes and businesses. Devices that receive infrequent security updates or rely on weak authentication may provide opportunities for malicious software to gain access and move between systems.

However, the researchers’ warning concerns a potential future capability. The report does not establish that AI worms have already achieved widespread autonomous propagation across consumer devices.

AI Advances Raise Broader Cryptographic Security Concerns

Anthropic’s warning comes amid separate concerns that increasingly capable AI systems could accelerate attacks against cryptographic systems used to protect digital assets.

Glassnode co-founder Rafael Schultze-Kraft estimated that approximately 6.26 million Bitcoin, representing close to one-third of the total supply, sit in addresses potentially vulnerable to attacks involving advances in AI or quantum computing.

Of that amount, around 4.33 million BTC are exposed because of address reuse, according to Schultze-Kraft. Moving those assets to fresh addresses could address that particular exposure, although it would not eliminate every potential cryptographic threat.

Ethereum researcher Justin Drake also warned that rapid advances in AI could affect the timeline for attacks against the elliptic curve digital signature algorithm (ECDSA), which is used to secure cryptocurrency wallets. Drake urged users to migrate to fresh wallets as a precaution.

These concerns remain distinct from Anthropic’s AI worm scenario. Self-replicating malware threatens device and network security, while attacks against cryptographic signatures would target the mathematical foundations used to authenticate transactions and protect private keys.

Crypto Industry Debates the Best Response

The potential risks to cryptocurrency security have prompted disagreement over whether moving assets to fresh addresses would provide sufficient protection against future cryptographic attacks.

Dragonfly managing partner Haseeb Qureshi criticized what he described as cryptographic doomerism, arguing that transferring tokens to new addresses is not a comprehensive solution to the underlying threat.

The disagreement reflects a broader challenge for the digital asset industry: preparing for emerging security risks without encouraging rushed actions that could introduce additional vulnerabilities.

For device manufacturers and cybersecurity teams, the AI worm warning similarly points to the importance of stronger security practices, regular updates and defenses that account for increasingly automated attacks.

Outlook

Anthropic’s warning highlights a potential new phase in cybersecurity, in which smaller AI models could automate parts of the process involved in spreading malicious software across connected devices. The timeline described by the researchers remains a forecast, and the report does not confirm that autonomous AI worms have reached the projected level of capability.

Further developments will depend on how quickly AI capabilities evolve and whether device manufacturers, security researchers and network operators can strengthen defenses before such systems become practical threats. For cryptocurrency users, the separate debate over AI-assisted cryptographic attacks reinforces the need to distinguish demonstrated vulnerabilities from future risks when evaluating wallet security.