In a moment that marks a new chapter in the geopolitics of intelligence, Anthropic has confirmed it detected and disrupted coordinated campaigns from Russian and Chinese state actors targeting its Claude AI system. The campaigns did not succeed, but their existence signals something profound: artificial intelligence has crossed a threshold, becoming not merely a commercial product but a strategic asset worth the attention of foreign intelligence services. What was once the domain of government networks and military infrastructure is now the territory of private AI laboratories, and the compani
Anthropic Disrupts Russian, Chinese AI Campaigns Targeting Claude Models
State actors have strong incentives to compromise them
Mark
So Anthropic found campaigns from Russia and China trying to compromise Claude. What exactly were they trying to do—steal the model, alter its outputs, what?
Mimi
The reporting doesn't specify the exact objective. It could have been espionage to understand how Claude works, or it could have been an attempt to inject malicious behavior into the system. Both are plausible targets for a state actor.
Luke
That's the thing—we know campaigns happened and were disrupted, but the actual mechanics are still opaque. Reuters doesn't say whether these were network intrusions, prompt injection attempts, or something else entirely.
Mark
Why would Russia and China care about Claude specifically? It's not like they control it.
Mimi
Because Claude is one of the most capable AI systems available outside their borders, and it's widely used. If you can compromise it, you gain leverage over a tool that millions of people and organizations depend on.
Luke
But we should be careful here. "Coordinated campaigns" is the language used, but we don't actually know the scale. Was this a sustained, resourced operation or a probing attempt? The reporting doesn't distinguish.
Mark
Did Anthropic say how they detected it?
Mimi
Not in detail. They identified it and disrupted it before significant damage occurred, but the specific detection method isn't disclosed.
Luke
Which makes sense from a security standpoint—you don't want to reveal your detection capabilities. But it also means we're taking Anthropic's characterization of the threat on faith.
Mark
So what happens now?
Mimi
Other AI companies are probably already scanning their own systems for similar activity. This is a signal that state actors see AI as a target worth investing in.
Luke
And Anthropic faces a real dilemma: disclose enough to help the industry defend itself, or stay quiet to protect operational security. There's no clean answer.
Le Pouls
- State-sponsored actors from Russia and China launched coordinated campaigns specifically designed to manipulate or breach Anthropic's Claude AI — not as opportunistic hackers, but as strategic adversaries with national interests at stake.
- The attacks represent a sharp escalation: foreign intelligence services are no longer content to target government systems alone, and are now treating commercial AI platforms as high-value geopolitical prizes.
- Anthropic's security team identified the coordinated, state-linked patterns of intrusion before significant damage occurred — a detection that required sophisticated monitoring capable of matching sophisticated attackers.
- By going public with the incident, Anthropic is sending a dual signal: confidence in its own defenses, and a warning to the broader AI industry that these threats are already in motion elsewhere.
- The disruption leaves urgent, unresolved questions — how much to disclose about attack methods, how to balance transparency with operational security, and how to prepare for the next wave that is almost certainly already underway.
In a moment that marks a new chapter in the geopolitics of intelligence, Anthropic has confirmed it detected and disrupted coordinated campaigns from Russian and Chinese state actors targeting its Claude AI system. The campaigns did not succeed, but their existence signals something profound: artificial intelligence has crossed a threshold, becoming not merely a commercial product but a strategic asset worth the attention of foreign intelligence services. What was once the domain of government networks and military infrastructure is now the territory of private AI laboratories, and the companies building these systems must now reckon with adversaries of a different order.
Anthropic, the AI safety company behind the Claude language model, has disrupted coordinated campaigns from Russia and China that were specifically designed to target and compromise its systems. The operations were detected before they succeeded in manipulating or breaching Anthropic's infrastructure — a disruption that required the kind of technical and intelligence work capable of identifying state-level coordination rather than isolated intrusions.
The campaigns mark a notable escalation in how state actors are thinking about artificial intelligence. Rather than targeting government networks or traditional critical infrastructure, these efforts focused on a private company's AI platform — a shift that reflects how central AI has become to national security calculations. Both Russian and Chinese actors appear to have recognized Claude's prominence and sought either unauthorized access to its systems or the ability to alter how the model behaves.
Anthropric's decision to disclose the incident publicly reflects a broader shift in how technology companies now handle security events. Where such breaches might once have been managed quietly, companies increasingly report them to warn the industry and establish a record of their security posture. The disclosure signals confidence in Anthropic's defenses while serving notice to other AI developers that similar threats are likely already targeting their systems.
The deeper tension the incident surfaces is structural: as AI systems grow more powerful and more central to economic and military advantage, state actors have strong incentives to compromise them. Claude has emerged as one of the most capable large language models outside of China, making it a natural target for foreign intelligence services. For Anthropic and the broader industry, the disruption raises urgent questions about what comes next — how much to reveal about how the attacks worked, and how to balance transparency against the risk of exposing vulnerabilities that other adversaries might exploit.
Anthropic, the AI safety company behind the Claude language model, has disrupted coordinated campaigns originating from Russia and China that were specifically designed to target and compromise its systems. The company detected the operations before they succeeded in manipulating or breaching its infrastructure, according to reporting from Reuters.
The campaigns represent a notable escalation in state-level interest in controlling or undermining commercial AI systems. Rather than targeting government networks or critical infrastructure in the traditional sense, these efforts focused on a private company's artificial intelligence platform—a shift that underscores how central AI has become to national security calculations. Both Russian and Chinese actors appear to have recognized Claude's prominence in the emerging AI landscape and sought to either gain unauthorized access to its systems or alter how the model behaves.
Anthropric's detection and disruption of these campaigns happened before significant damage occurred. The company's security team identified the coordinated nature of the attacks, meaning they recognized patterns suggesting state-level coordination rather than isolated intrusions. This kind of attribution—determining that campaigns are linked and state-sponsored—typically requires substantial technical evidence and intelligence work. The fact that Anthropic could identify and stop the operations suggests the company maintains security monitoring capabilities sophisticated enough to catch sophisticated state actors in the act.
The timing of this disclosure reflects a broader shift in how technology companies now handle security incidents. Where once such breaches might have been handled quietly, companies increasingly report them publicly, both to warn the broader industry and to establish a record of their security posture. Anthropic's decision to make this public signals confidence in its defensive measures while also serving notice to other AI developers that similar threats are likely already in motion against their systems.
The incident highlights a fundamental tension in the current moment: as AI systems become more powerful and more central to economic and military advantage, state actors have strong incentives to compromise them. Claude has emerged as one of the most capable and widely used large language models outside of China, making it a natural target for foreign intelligence services seeking either to understand its capabilities or to degrade its performance. Russia and China both have documented histories of targeting American technology companies, but this appears to be among the first public instances of coordinated state campaigns specifically aimed at manipulating a commercial AI system.
For Anthropic and other AI companies, the disruption of these campaigns raises urgent questions about what comes next. If Russia and China are already attempting to compromise Claude, similar efforts are almost certainly underway against other major AI systems. The company will likely face pressure to disclose more details about how the attacks worked and what specific vulnerabilities were exploited—information that could help the broader industry harden its defenses, but that could also reveal gaps that other actors might exploit. The balance between transparency and operational security has become one of the defining challenges for AI companies operating in a geopolitically fractious environment.
Envie de l'histoire complète ?
Lire l'original sur Reuters ↗
Technology