Bitget Wallet expands AI phishing protections as weekly detections hit 37,000

What the company's numbers show

The wallet logged about 37,000 phishing-related signing risk detections per week in Q3, up from 5,000 in Q2 (those figures come from Bitget Wallet's own data). Phishing-related signing risks made up 57.4% of detections across four monitored categories in Q3. Malicious addresses followed at 16.5%, then risky contracts at 14.2% and malicious approvals at 11.8%.

There's a catch. Kan said the company introduced AI-assisted monitoring in Q2 and expanded its detection capabilities gradually after that, so better tooling and more attacks are tangled together in the same chart.

“The data does not allow us to quantify how much of the increase came from actual attack activity versus improved detection,” Kan said.

How GetShield works

The company expanded AI-assisted analysis built into GetShield, its proprietary security engine, which examines website code, domain characteristics and similarity to legitimate sites. It also simulates transactions and checks signing permissions to catch hidden approvals and unexpected asset transfers before a user authorizes anything. Kan said the system pairs AI analysis with rule-based screening and human review.

According to Bitget Wallet, its AI helped identify 90.2% of newly blocked risky domains in the first four weeks of Q4, up from 70.2% in Q3 and 36.3% in Q2. The company said nearly 70% of those domains had no matching alerts in external threat intelligence databases. Kan added that this doesn't necessarily mean those providers missed the threats, because detection criteria and listing times vary.

What isn't disclosed

The wallet didn't disclose its false positive rate. It said the system can warn users, ask for extra confirmation or block an interaction depending on risk level, and anyone who thinks a site or DApp was wrongly flagged can contact support for a security review. Kan said detection doesn't upload private keys, recovery phrases or sensitive personal data. The company didn't say whether its AI analysis runs on users' devices or on its servers, or whether third-party language models are involved.

The expansion comes as attackers lean on AI to generate convincing phishing websites and deceptive content. Kan said those tools let scammers develop and test scams at scale. Bitget Wallet also cited Chainalysis research showing that scams linked to AI vendors generated 4.5 times more revenue per operation than scams without such links.

Kan also said no detection system can catch every emerging threat.

Bitget Wallet said it plans to expand AI-driven threat detection further and explore AI-assisted incident response under human oversight.