TechRadar News.
Technology

Stolen Access Key Connects Fireworks AI Leak to Compromise of Cursor’s Claude Code Extension

Stolen Access Key Connects Fireworks AI Leak to Compromise of Cursor’s Claude Code Extension

Investigators have linked the latest breach of Anthropic’s Claude Code add‑on in Cursor’s marketplace to an access key that was stolen in an earlier Fireworks AI intrusion. That earlier incident revealed internal credentials, enabling a threat actor to acquire a key that Cursor had earlier granted to the Fireworks service for integration.

Cursor’s own statement says the adversary leveraged the stolen key to breach the Claude Code extension—a utility that lets programmers produce and modify code using Anthropic’s AI model inside the Cursor IDE. The release states, "An attacker obtained an access key we had issued to Fireworks," verifying the connection between the two events.

This case underscores how tightly linked AI‑tool ecosystems can turn shared authentication into a single point of failure. Anthropic, the maker of Claude, has not revealed how much data may have been exposed, but warned that the compromised extension might have granted unauthorized parties access to user prompts and code fragments processed by the service.

Cursor, a developer‑centric marketplace for extensions, has temporarily taken down the Claude Code add‑on pending a forensic investigation. The firm says it is revoking the stolen key, rotating every credential linked to third‑party integrations, and tightening key‑distribution policies to avert future incidents.

Fireworks AI, a startup delivering generative visual‑content tools, experienced a breach earlier this month that revealed internal systems and multiple API keys. Although the attack did not specifically aim at Cursor, the exposed key seems to have been reused by attackers to penetrate the wider AI‑extension ecosystem.

Analysts point out that the episode highlights the need for strong secret‑management, particularly as AI services grow more modular and depend on cross‑platform authentication. Specialists advise the use of short‑lived tokens, zero‑trust network designs, and ongoing monitoring of credential activity to lessen the impact of inevitable future leaks.

Anthropic and Cursor have both committed to informing impacted users and offering advice on securing their accounts. As probes proceed, the case serves as a reminder that the swift growth of AI‑powered developer tools imposes greater security duties on both providers and users alike.

Source: feedburner
TechRadar Desk — Editorial desk.

Comments (0)

Be the first to comment.

Join the discussion

Protected by reCAPTCHA v3

Related