Secure GenAI
Secure GenAI Podcast
1st AI report, DeepSeek, Gmail hack, new code practices, US treasury vendor details.
0:00
Current time: 0:00 / Total time: -6:19
-6:19

1st AI report, DeepSeek, Gmail hack, new code practices, US treasury vendor details.

GenAI Safety & Security | Jan 27 - Feb 3, 2025

Notice: FY2024 is coming in Feb 2025.

Highlights

🌎 AI international report: 100 experts.

🚨 DeepSeek's AI raises safety concerns.

📧 Gmail users targeted with AI hacks.

🛡️ New AI cybersecurity code of practice.

💰 US Treasury vendor breach details.

Secure GenAI is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.

Deep Dive

🌎 1. First International AI Safety Report - AI Action Summit GOV.UK

  • Report is a global handbook for AI.

  • 100 AI experts from 30 countries.

  • AI agents can now perform complex task

  • AI risk depend on policy decisions

  • Key is to understand AI's internals.

Notice: the 1st official report following several international AI safety summits.

🚨 2. Database Leaking Sensitive Info, Including Chat History Wiz Blog, PCMag

  • Over 1 million lines of log exposed.

  • API keys, chat histories leaked.

  • Full database control was accessible.

  • Unsecured ClickHouse database found.

  • Security overlooked due to AI growth.

  • DeepSeek R1 failed all safety tests.

  • 100% success rate for harmful prompts.

  • Lacks proper defense compared to others

  • Low development budget a potential cause.

  • Strong content restrictions on China topics.

Mitigation: Use tools to test internal AI model. Set guidelines for internal and external model use.

🚨 Privacy EU complaint The Brussels Times, Euroconsumers

  • DeepSeek hit 2.6 million downloads quickly.

  • GDPR breaches are serious concern.

  • Italian data authority blocked DeepSeek app.

  • Multiple EU countries launched investigations.

  • Euroconsumers drives legal action for users.

  • Data transfer to China is concern.

  • Lacks transparency for user data use.

  • No minor data safety measure taken.

  • GDPR violations on data processing found.

  • Temporary restriction of service asked.

Mitigation: Review data governance and transparency in data handling.

🚨 Database cyberattack Forbes, CSO Online

  • Misconfigured cloud storage cause of breach.

  • Leaked data can be used for cyberattacks.

  • User data, keys, and secrets leaked.

  • Proprietary AI models were potentially exposed.

  • Full admin control was accessible.

  • DeepSeek secured database after notification.

  • Lack of security is a large issue.

    Secure GenAI is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.

📧 3. Gmail 2.5 Billion Users—AI Hack Confirmed Forbes

  • AI-powered phishing attacks detected.

  • Hackers use fake Google tech support.

  • Attackers sending Gmail reset codes.

  • Advanced Protection program provides security.

  • Attack was hard to detect by victims.

Mitigation: Educate users; enable Advanced Protection. Consider passkeys for verification. Review all your device IPs on daily or weekly basis.

🛡️4. Code of Practice for the Cyber Security of AI GOV.UK

  • Code for AI security by UK government.

  • Addendum to the Software Code of Practice.

  • Covers five AI lifecycle phases.

  • Guidance for developers and operators.

  • Focus on threat assessment and response.

Mitigation: Review and implement AI security framework. Educate developers on AI-specific risk management.

💰 5. US Treasury vendor breach investigation BeyondTrust

  • 17 Remote Support SaaS users were impacted.

  • API key was compromised via ZERO-day.

  • China-nexus threat actors involved.

  • Remote Support patched after the incident.

  • Investigation completed Jan 17th.

Mitigation: Stay current on releases and implement patches. Review authentication security measures.

Thanks for reading Secure GenAI ! This post is public so feel free to share it.

Share


Discussion about this podcast