Nansen, a company specialized in blockchain analysis, took a new step in the use of artificial intelligence on Thursday (25) by presenting Nansen AI, an AI agent developed from its vast database with wallets identified across more than 20 blockchains.

The feature was built on the Claude language model from Anthropic, but gained a differential: integration with Nansen's proprietary data. This way, instead of relying only on traditional dashboards and charts, users can interact through a conversational interface to investigate market trends and wallet movements.

At launch, the solution was presented as a kind of research assistant. With it, it is possible to request the identification of trade signals, ask for explanations about asset flows, or track operations of investors considered 'smart money.'

The project, however, aims higher: Nansen plans to eventually include the ability to execute trades directly through the tool — always with user confirmation. To attract new customers, the company has reduced the monthly subscription price from $99 to $69.

What the AI agent does — and what it doesn't do yet

According to Nansen, the strength of its AI agent lies in data advantage: over 500 million labeled addresses provide context of identity and behavior to the model's predictions. Due to this specialized input, the company claims that the agent outperforms general-use models like ChatGPT or Grok in specific cryptocurrency prediction tasks.

Currently, the agent supports portfolio context (such as Ethereum wallets and EVM-compatible networks). Trade execution will be added later: when activated, the agent will be able to propose trades but will need user confirmation before sending any transaction. Nansen wants to validate this 'central cycle' before allowing autonomous flows.

Despite the enthusiasm at launch, Nansen has not published a technical white paper. There is still no public disclosure regarding the agent's accuracy, false positive rate, robustness, or testing against attacks. This lack of transparency raises the question: is this mainly a marketing play, rather than a scientific delivery?

Risks and challenges

Nansen's move into AI carries embedded risks — especially in the financial context. A recent academic paper, 'AI Agents in Cryptoland: Practical Attacks and No Silver Bullet' – warns about context manipulation, where attackers alter the history of prompts or memory to induce the agent into harmful actions or incorrect predictions.

Agent-based trading systems need to protect themselves against hallucinations and unauthorized executions — especially in a volatile environment like cryptocurrencies. Nansen's commitment to keeping human confirmation in the loop is a protective measure, but it has yet to be tested in high-speed markets.

Another challenge is the bias or aging of data. The value of labeled addresses diminishes over time; if the bot's guidance is based on outdated patterns, users may be misled. And, since the model's results are still not transparent, the ability to audit or verify predictions independently is limited.

Why does this matter

If Nansen AI really provides reliable insights faster than chart analysis, it could lower the entry barrier in cryptocurrency trading. A user asking 'Which EVM wallets are accumulating this token today?' and receiving an instant answer gains power even without being an expert.

This also signals a broader shift: analytics providers are transforming into agent platforms. But, to move beyond just a flashy demonstration, Nansen AI will need to prove that its predictions hold up in live markets — and that it withstands adversarial stresses.

In the relentless crypto world, many AI agent projects have already failed when real money came into play.