Anthropic Building Tool to Expose Chinese Models Secretly Using Claude

Anthropic is developing an open-source tool that could expose AI firms secretly using Claude to generate responses while claiming the answers came from their own models.

Key points

• According to Anthropic's September 2026 report, Moonshot AI secretly forwarded almost 300, 000 Kimi customer requests to Claude over ten days, primarily using its Opus models.
• Earlier, in February, Anthropic accused DeepSeek, Moonshot, and MiniMax of conducting large-scale model-distillation campaigns involving more than 16 million exchanges through approximately 24, 000 fraudulent accounts.
• In September 2026, Anthropic accused Moonshot AI of secretly routing Kimi users' requests through Claude and presenting the responses as Kimi's own.
• Anthropic alleged that Moonshot accessed Claude through 5, 380 fraudulent accounts, mostly appearing to operate from Singapore and Japan.
• Using four-bit quantized weights and a single Nvidia H100 GPU, the system generated proofs at approximately 110 tokens per second.

For context, the system, called FlashProve, uses cryptographic technology to verify which AI model produced a response. Although Anthropic has not explicitly linked the project to Chinese AI firms, it could assist address practices the firm has previously accused Chinese competitors of using.

In September 2026, Anthropic accused Moonshot AI of secretly routing Kimi users' requests through Claude and presenting the responses as Kimi's own. The firm has additionally documented multiple cases of Chinese AI laboratories allegedly extracting Claude's capabilities to train competing models.

Nevertheless, FlashProve is designed to verify which model generated an output, not determine whether another model was trained using Claude's responses. Chinese AI Firms Accused.

Anthropic alleged that Moonshot accessed Claude through 5, 380 fraudulent accounts, mostly appearing to operate from Singapore and Japan. It additionally accused the business of saving conversations and extracting Claude's reasoning data to improve its own models.

Model distillation involves using responses from a more capable AI system to train another model. Although the technique has legitimate applications, Anthropic notes unauthorized extraction of Claude's capabilities violates its terms.

In September, Anthropic additionally accused Alibaba, Zhipu, and Xiaomi of similar activities and alleged that DeepSeek had secretly forwarded some customer requests to Claude. These remain Anthropic's allegations rather than independently established findings. Anthropic is Regarding to Achieve Self Building AI. How Anthropic's FlashProve Works.

As a system for provable inference, allowing users to mathematically verify which AI model generated a response, anthropic introduced FlashProve on November 2, 2026.

At present, users generally have to trust businesses when they claim a particular model produced an answer.

For example, an AI provider could secretly replace an expensive model with a cheaper alternative to reduce costs. Alternatively, it could employ a more powerful model during benchmark testing to produce its own system appear better. FlashProve intends to prevent such misrepresentation through cryptographic verification.

To employ the system, AI businesses would publish a unique cryptographic fingerprint associated with each model.

When the model generates a response, the provider can produce mathematical proof linking that output to the published fingerprint. Users can then verify whether the response came from the advertised model. If a different model generated the response, the verification should fail.

Meanwhile, the approach is similar to cryptographic email signatures. It support confirm that messages came from their maintained senders. AI Firms Won't Need to Reveal Their Models.

One major advantage of FlashProve is that firms can verify their models without revealing their internal weights.

Model weights are the numerical values that determine how an AI system processes information and generates responses. Firms such as Anthropic and OpenAI generally keep the weights of their leading models private.

FlashProve allows providers to publish cryptographic fingerprints without exposing this proprietary information.

Although generating the proofs requires considerably more computing power, the system is additionally designed to backing lightweight verification on local devices.

Participation would require AI providers to publish the necessary fingerprints and generate proofs. FlashProve Could Additionally Detect Misleading AI Claims.

Beyond identifying which model generated a response, FlashProve could support verify claims produced by firms using AI for professional tasks.

While another firm might advertise that a powerful AI system reviewed its business plan, for example, a developer might claim that an advanced cybersecurity model scanned its software for vulnerabilities. FlashProve could provide mathematical evidence confirming which model performed the task.

It could additionally assist researchers verify the models applied during benchmark testing, making it harder for businesses to secretly substitute more capable systems.

Nevertheless, proving a model's identity does not guarantee that its answers or conclusions are accurate. Anthropic Tests FlashProve on a 400-Billion-Parameter Model.

Anthropic has demonstrated FlashProve using Meta's Llama 4 Maverick. It contains 400 billion total parameters.

For a request containing 1, 024 tokens, the proof was approximately 19MB and could be verified on a CPU within seconds.

Nevertheless, proof generation required approximately 70 to 95 times more GPU computation than comparable inference using vLLM. This extra computing cost remains a major obstacle to widespread adoption. FlashProve Is Open Source but Still Experimental.

Anthropic has published FlashProve's technical specification, mathematical research, and working code on GitHub under an open-source license.

Notably, the firm states the project is still in development and has limitations involving model computations, backed output types, and the relationship between its optimized implementation and formally verified protocol.

Anthropic has not confirmed aims to deploy FlashProve in Claude or any other commercial AI service.

Instead, researchers are making the technology available so others can study, test, and improve it.

If successfully developed and adopted, FlashProve could create AI businesses more accountable for claims regarding which models power their services.

Nevertheless, it would not independently prove unauthorized model distillation or identify the training data employed to build competing AI systems. Stay Connected with ProPakistani.

Obtain the latest tech news, telecom insights, and product launches wherever you prefer. Follow on Google Discover. Follow on Google News Join WhatsApp. See more ProPakistani stories in Google Search and Top Stories.

Technology and Automotive Specialist covering the latest cars, smartphones, AI breakthroughs, and.

Taken together, the developments around anthropic Building Tool to Expose Chinese Models Secretly Using Claude point to a situation that is still moving, and the coming days should bring more clarity.