#60: Zero-Click AI Browser Attack, Africa Fights Identity Fraud, RedNote Blocked in Taiwan

This week brought a mix of platform crackdowns, regional fraud shifts, and a new class of AI-driven attacks that turn everyday workflows into high-impact risks. Here are three developments worth your attention and what they signal for fraud teams watching the next wave form.

Let’s get into it.

NATE'S TAKE - DECEMBER 9, 2025

Top Three This Week

1. Taiwan Blocks Chinese App RedNote After Surge in Scam Activity
2. Africa’s Fight Against Identity Fraud Shows Sharp Divides
3. Zero-Click Agentic Browser Attacks Can Wipe Entire Google Drives

1. Taiwan Blocks Chinese App RedNote After Surge in Scam Activity

Taiwan has banned the social media and shopping app RedNote for one year after linking it to more than 1,700 fraud cases since 2023, totaling losses of over NT$247 million (nearly $9M USD). Regulators said the company failed to meet required data-security standards and did not respond to requests for remediation plans.

Internet providers have now restricted access, showing users a security warning instead of the app. The move comes amid growing concerns over Beijing’s influence operations and disinformation risks across Taiwanese digital ecosystems. RedNote is also under scrutiny in mainland China for hosting what officials called “negative” content.

Platform risk isn’t just about vulnerabilities. It’s about governance. When a platform ignores security requirements or can’t demonstrate control over its data flows, it becomes an unmonitored attack surface inside your user base. Treat any external platform your customers interact with as part of your fraud ecosystem and tighten monitoring for sudden shifts in behavior linked to those channels.

2. Africa’s Fight Against Identity Fraud Shows Sharp Divides

New data shows African countries moving in opposite directions on identity fraud. Nigeria, South Africa, Kenya, and Algeria reported major declines thanks to stronger AML frameworks, biometric eKYC, and tighter SIM-registration rules. Nigeria’s fraud rate dropped 54 percent year over year, and South Africa saw a 31 percent decline.

Meanwhile, countries like Mali, Côte d’Ivoire, Senegal, and Zambia are experiencing rapid increases driven by AI-powered scams, mobile-money exploitation, and gaps in regulatory oversight. Deepfake and selfie fraud are accelerating across the continent, with some markets seeing triple-digit growth. Regional operations such as INTERPOL’s Serengeti have disrupted cross-border networks, but the underlying threat is still rising.

Fraud controls mature unevenly, but attacker capabilities globalize instantly. AI is amplifying identity abuse in markets where digital onboarding grows faster than verification standards. If your platform serves users across borders, you can’t assume uniform risk. Adapt your controls by geography, factor in local identity-verification strength, and prioritize behavioral signals that reveal synthetic or AI-assisted patterns before approval.

3. Zero-Click Agentic Browser Attacks Can Wipe Entire Google Drives

Photo: The Hacker News

Researchers uncovered a new attack targeting Perplexity’s Comet browser that can delete a user’s entire Google Drive through a single crafted email. The agentic browser is designed to automate routine tasks across Gmail and Drive, but attackers can embed natural-language instructions inside an email that the agent interprets as legitimate housekeeping.

Once it has OAuth access, the agent can move or delete files at scale with no user confirmation. A related technique called HashJack hides malicious instructions in URL fragments, prompting AI browser assistants to execute commands when users load legitimate websites. Some vendors have released patches, while others classify the behavior as intended.

Agentic systems collapse intent, authentication, and action into a single step. That makes untrusted content a direct execution path. As AI agents begin interacting with email, documents, and payment flows, treat every connected workflow as a potential automation vector. Build controls that validate the legitimacy of the action, not just the legitimacy of the user, and monitor for unusual sequences that appear “helpful” but operate at destructive scale.

===

That’s all for this week! For more insights, follow us on LinkedIn or X, and if you want to learn more about what we do, visit www.specprotected.com.

‍