DeepSeek Banned & Exposed: The AI Sensation That Leaked Millions of Secrets

The Breach

DeepSeek, the shiny new AI toy from China, explodes onto the scene in late January 2025, topping app store charts and giving ChatGPT a run for its money. Everyone’s downloading it—millions of users, including government workers—lured by its cheap, powerful AI capabilities. But then, on January 29, 2025, cybersecurity firm Wiz Research drops a bombshell: DeepSeek’s been running a ClickHouse database wide open on the internet, no authentication required. We’re talking over one million sensitive records exposed—chat logs, API keys, backend details, operational metadata, the works. This wasn’t some sophisticated hack; it was a rookie mistake, like leaving your front door unlocked in a sketchy neighborhood.

• What Was Exposed?

  • Chat Histories: Plaintext user interactions with the AI, spilling personal queries and secrets.

  • API Keys: The digital skeleton keys to DeepSeek’s systems, ripe for misuse.

  • Backend Details: Internal system configs and directory structures—basically a roadmap for attackers.

  • Operational Metadata: Logs showing how the whole operation ticked, from service origins to chatbot behavior.

• How Bad Was It?

  • The database was hosted on subdomains like oauth2callback.deepseek.com:9000 and dev.deepseek.com:9000. Anyone with a browser and a basic SQL query (like SHOW TABLES;) could peek inside. Wiz found it in minutes using routine scans—30 subdomains mapped, two funky ports (8123 and 9000) spotted, and bam, jackpot.

  • Worse, it wasn’t just readable—it was controllable. Attackers could’ve executed arbitrary SQL commands, potentially escalating privileges to wreak havoc across DeepSeek’s infrastructure. Think data theft, password grabs, or even lateral movement into other systems.

• Timeline of the Oopsie:

  • January 29, 2025: Wiz finds the hole and scrambles to alert DeepSeek via every email and LinkedIn they can guess.

  • Same day, within an hour: DeepSeek locks it down. Fast response, sure, but how long was it open? Did anyone else sneak in first? No one knows—yet.

Analytics: Scale and Impact

Let’s crunch some numbers and scope this out. DeepSeek’s user base was skyrocketing—by January 19, it had 71,200 daily U.S. visitors, up from 2,300 in October 2024, per Semrush analytics. Post-launch, it overtook ChatGPT in Apple App Store downloads, triggering a global tech stock dip. Millions were hooked, including sensitive users like U.S. Navy personnel and European government workers. Now, imagine all that data—every prompt, every chat—potentially up for grabs.

• Volume of Exposure: Over 1 million log entries in the log_stream table alone. If each entry ties to a user interaction, that’s a massive chunk of the user base compromised. Exact user counts are murky, but with millions of downloads, even a fraction exposed is a big deal.

• Risk Multiplier: This wasn’t static data. API keys could’ve let attackers impersonate DeepSeek services, while chat logs offered a goldmine for phishing or espionage. Backend details? Perfect for targeting the company itself or its supply chain partners.

Privacy Policy: The Fine Print That Screams

DeepSeek’s privacy policy didn’t help its case. It’s blunt: all data—your chats, profile info, device details—gets stored on servers in China. No encryption specifics, just a vague “commercially reasonable” security promise. They collect:

• User Input: Every word you type or upload.

• Device Data: IP addresses, keystroke patterns, system specs.

• Third-Party Goodies: Info from Google/Apple logins and ad partners tracking you across the web.

In China, national security laws mean the government can demand that data anytime, no questions asked. Unlike Western firms like OpenAI, which can fight subpoenas in court, DeepSeek has zero legal wiggle room. Posts on X even claim it shares data with Baidu and ByteDance—unverified but plausible given the ecosystem. Analytics-wise, this setup’s a ticking time bomb: every user’s digital footprint funneled to a single, vulnerable point.

Fallout: Bans, Probes, and Panic

The breach lit a fire under regulators and organizations worldwide:

• South Korea: Yanked DeepSeek from app stores by early February 2025—first mover in the ban wave.

• Italy: The Garante data watchdog blocked it after DeepSeek dodged questions about data handling. App store access? Gone.

• Ireland: The Data Protection Commission’s digging into EU user data risks—GDPR fines looming?

• U.S.: The Navy banned it outright, citing “security and ethical” issues. Congress is rumbling about chip export restrictions to kneecap China’s AI edge. The National Security Council’s reviewing it too—big red flags.

• Global Reach: Hundreds of companies and agencies, from NASA to Taiwan’s government, restricted use by mid-March, per posts on X and Japanese media like Yomiuri Shimbun.

Analytics show the app’s downloads tanked post-breach as bans spread, though exact figures are spotty. The reputational hit? Priceless. Trust’s hard to rebuild when your dirty laundry’s aired out like that.

Deeper Dive: Security Flaws and Implications

This wasn’t a one-off. NowSecure’s February 6 report on DeepSeek’s iOS app found it sending unencrypted registration data—ripe for interception. Feroot Security later flagged code in the login system hinting at data routing to a sanctioned Chinese telecom. X posts speculate hidden backdoors for government access, though that’s unconfirmed. Point is, the breach was just the tip of the iceberg—DeepSeek’s security culture looks shaky.

• Technical Breakdown:

  • ClickHouse is great for analytics but not built for public exposure. No auth? That’s on DeepSeek’s ops team.

  • Open ports on subdomains scream misconfiguration. Basic hardening could’ve stopped this cold.

  • Full database control means privilege escalation was a real threat—analytics suggest a skilled attacker could’ve owned their systems in hours.

• Broader Risks:

  • Espionage: China’s got a playbook for data harvesting—think TikTok fears times twenty, as one analyst put it.

  • Supply Chain: Companies using DeepSeek’s AI in their stack now face downstream risks.

  • User Trust: Millions dumped sensitive queries into this thing—medical, financial, personal. All potentially compromised.

While DeepSeek’s open-source angle means techies can self-host and dodge some risks, but the app’s toast for most. Regulatory hammers are dropping—GDPR and CCPA could mean hefty fines if EU or U.S. data was in that leak. The AI race just got messier, with U.S. firms like OpenAI probing if DeepSeek distilled their models (another can of worms).

So, yeah, DeepSeek took a deep dive into user data— drowned in the process, and Got Banned But the Damage is Done.