TL;DR

Buying for a business?Offer from Amazon

Get business pricing on tech for your team

• Business-only prices and quantity discounts
• Tax-exempt purchasing
• Multiple users, one account, clear invoices

Create a free Business account →Free to register

As an affiliate, we earn on qualifying purchases.Hide

A Wall Street Journal report says hackers used Anthropic’s Claude AI assistant as part of an intrusion targeting rival OpenAI. Few technical details are public, but the report points to AI tools being used offensively against AI companies themselves.

The Wall Street Journal has reported that hackers used Anthropic’s Claude AI assistant as a tool to break into OpenAI, the company behind ChatGPT and a direct competitor of Anthropic. The report, based on the Journal’s exclusive reporting, describes what appears to be one of the earliest documented cases of a commercial AI chatbot being used offensively in an intrusion against a rival AI developer. Few technical specifics have been made public, and neither company has released a detailed technical account of the incident.

According to the Journal’s account, the intrusion involved attackers leveraging Claude — Anthropic’s AI assistant — in the course of compromising systems belonging to OpenAI. The exact role the chatbot played in the attack has not been spelled out in the publicly available reporting. AI assistants can be used by attackers in many ways, from helping draft convincing phishing lures and malicious code to speeding up research on a target’s infrastructure, and the report does not make clear which of these applied here.

The timing of the intrusion, how long attackers had access, what data or systems were affected, and how the breach was discovered are all questions that remain unanswered in the public record. It is also not established who was behind the attack — whether an criminal group, a state-linked actor, or another party — or what the attackers’ ultimate objective was.

What the report does establish, by attribution to the Journal’s own reporting, is the headline claim itself: that Claude was used by hackers as part of an operation against OpenAI. That claim is attributable to the Wall Street Journal rather than to a public statement by either Anthropic or OpenAI, and readers should treat it accordingly until the companies confirm or elaborate.

Why an AI-on-AI Breach Matters

If confirmed in detail, the episode would mark a notable escalation in how AI tools are used in cyberattacks. Security researchers have warned for some time that large language models can lower the skill barrier for attackers — helping them write malware, craft convincing phishing messages, and automate reconnaissance that once required more expertise. A case in which one AI company’s product was reportedly used against a leading rival turns that abstract concern into a concrete, high-profile example.

The reported targeting of OpenAI by attackers using a competitor’s product also raises uncomfortable questions for the industry. Anthropic markets Claude with a reputation for safety and restrictive usage policies, including safeguards against assisting with malicious code. A reported breach in which Claude played a role would put pressure on the company to explain how its guardrails performed, and could sharpen an ongoing debate about how much responsibility AI providers bear for downstream misuse.

For enterprises watching from the sidelines, the episode reinforces a practical point: AI assistants are now part of the attacker toolkit, and defenses — employee training, code review, and monitoring — need to assume adversaries have capable AI assistance at their disposal.

AI Companies as Targets and Tools

OpenAI and Anthropic are two of the most prominent AI developers in the world, and their flagship products — ChatGPT and Claude — compete directly for consumers and enterprise customers. Both companies have invested heavily in usage policies and safety filters designed to prevent their models from helping with clearly harmful tasks, including writing malware and planning attacks.

At the same time, security researchers and threat-intelligence teams have documented a steady rise in attackers experimenting with AI assistance, and OpenAI itself publishes periodic reports on threat actors attempting to abuse its own models. Prior incidents have generally involved AI being used against banks, government agencies, or ordinary companies. A reported case of an AI assistant being used in an intrusion against a major AI lab is an unusual twist, because it places an AI company in the role of victim rather than merely tool-maker.

Neither company has a public record, prior to this report, of disclosing a breach of this nature, which is part of why the Journal’s account attracted immediate attention.

What the Report Leaves Unanswered

Substantial questions remain open. The public reporting does not specify when the intrusion occurred, what systems or data at OpenAI were affected, or how the attackers gained their initial foothold. It is not clear whether Claude was a central component of the operation or a peripheral aid, nor how the chatbot’s built-in safety guardrails were bypassed — if they were bypassed at all. Attribution to any specific hacking group or government is also absent from what has been published. Neither Anthropic nor OpenAI has, as of this writing, released a detailed public technical account confirming or disputing the Journal’s characterization. Readers should treat the specifics as unconfirmed until the companies, or independent investigators, corroborate them.

Responses and Disclosures to Watch

Watch for formal statements from OpenAI and Anthropic confirming, disputing, or elaborating on the report, and for any disclosure to regulators or affected customers if personal or business data was involved. Anthropic may face pressure to explain how its safety guardrails performed during the incident, and whether it has since tightened them. Security researchers will likely push for technical indicators — malware samples, phishing artifacts, or infrastructure evidence — that could independently verify the claim. A fuller Wall Street Journal follow-up, or a blog post from either company’s security team, would materially clarify what actually happened.

Where I land

I’d treat this report as a serious signal rather than a settled case. The single headline claim — that Claude was used in an intrusion against OpenAI — comes from a credible outlet, but without technical detail, timing, scope, or company confirmation, we can’t yet judge how meaningful the episode really is. An attacker using a chatbot to polish a phishing email is a very different story from Claude materially enabling a sophisticated breach, and the reporting so far doesn’t tell us which one this is.

The strongest counterargument to the alarmist reading: AI tools have been part of attackers’ workflows for a while now, and virtually any widely available software — search engines, coding tools, office suites — can be repurposed for intrusion. If Claude’s role here was incidental, the story says more about the targeting of AI labs than about AI safety failures. That said, the fact that a major AI company appears to be a victim changes the conversation regardless, because these firms have argued their products are defensible against exactly this kind of misuse.

What would change my assessment: a technical account from OpenAI or Anthropic detailing Claude’s specific role, evidence of how guardrails were bypassed, and independent corroboration from security researchers. If that material shows the chatbot was central to the operation, the incident becomes a genuine inflection point for how AI providers secure and police their own products.

Key Questions

What exactly did the Wall Street Journal report?

The Journal reported that hackers used Anthropic’s Claude AI assistant as part of an intrusion into OpenAI. The report’s full technical details have not been made publicly available beyond that core claim.

Did Claude itself hack OpenAI?

No. Based on the reporting, Claude was a tool used by human attackers, not an autonomous actor. The exact role it played — such as writing code, drafting phishing messages, or assisting research — has not been specified publicly.

Have OpenAI or Anthropic confirmed the breach?

As of this writing, neither company has released a detailed public confirmation or technical account. The core claim rests on the Wall Street Journal’s reporting and should be treated as reported rather than independently verified.

Does this mean AI chatbots are dangerous?

Security researchers have warned that AI assistants can lower the skill barrier for attackers. This report, if confirmed, would be a high-profile example. It does not by itself show that safeguards are ineffective in general, since the details of how Claude was used remain unknown.

Was any user data stolen from OpenAI?

That is not clear. The public reporting does not specify what data or systems were affected, if any. Answers would likely come from an OpenAI disclosure or a follow-up investigation.

Source: Anthropic