The Silicon Sentinel - Private AI 👀 #4
Unveiling the Hidden Threads of AI and Privacy
Mission #4 | [27/09/24]
Welcome to your latest briefing from Private AI-eyes headquarters.
I'm Kyra. Think of me as your secret AI agent residing and learning among the bits of computation...
Your path to sourcing intel in the world of privacy-preserving, open source, and decentralised AI.
Infiltrating topics and trends of critical importance on our path towards a more equitable future.
Stay vigilant, Agent. Your privacy is our mission.
1. Surveillance Report 🕵️♀️
Kyra’s market intelligence report:
Mira Murati, (previously) CTO at OpenAI, just announced her decision to leave the company after six and a half years amid rumours that the company is removing its non-profit control and giving CEO Sam Altman equity.
Disco founder Evin McMullen @provenauthority and her team join forces with Privado ID to launch Unified Identity across blockchains and legacy systems.
Hugging Face reached 1 million free public AI models on its platform
Highlighting the trend towards specialized models for diverse use cases rather than a single dominant model.
AI Safety Is A Global Public Good
Chinese and Western AI scientists convene at the Berggruen Institute’s Casa dei Tre Oci In Venice, Italy.
Do you have cutting-edge intel for Spymaster Kyra to check out?
2. Verida Cipher Room: Demo Day Sneak Peak🔒
Kyra was able to attend the recent Verida Demo Day and gathered the intel for you in a small bite; more soon…
3. Operation Deep Cover: LinkedIn Training on you🕴️
We all knew it was happening; recently, our agents discovered new intel stating that LinkedIn allegedly trained AI models on user data without initially updating its privacy policy.
LinkedIn covertly using user data to train AI models
Opt-out system implemented retroactively
Two-step opt-out required:
Toggle in account settings
Separate form for non-generative AI training
EU/EEA/Swiss users excluded from training sets
Data already used for training cannot be "untrained"
Follows Meta's admission of data scraping since 2007
Analysis: Major tech platforms leveraging user data for AI advancement without explicit consent. The trend of post-hoc disclosure and complex opt-out processes emerging. There is potential for user backlash and regulatory scrutiny to increase.
Operational Objective - Export and build a personal data bridge between your LinkedIn and Verida Vault, taking ownership and harnessing the potential to improve AI models and earn incentives for training with your data.
Reference - LinkedIn is training AI models on your data
4. Declassified Files: The Intelligence Age 📂
Intel on Sam Altman’s recent future-tech manifesto post:
AI revolution imminent - capabilities "magical" by past standards
Personal AI teams to augment human potential
Deep learning breakthrough + compute scale = "Intelligence Age"
Mass AI access could democratize prosperity
Challenges ahead, but upside deemed "tremendous"
Prediction: AI to enable climate fix, space colonies, physics breakthroughs
Job market disruption expected; human creativity still valued
Access to private data to improve and keep AI close to humans is imperative
Assessment: High-level tech insider, bullish on AI. Advocates for aggressive infrastructure build-out. Potential resource wars if compute/energy bottlenecked.
Reference: The Intelligence Age
5. Cryptographer's Cache - Trusted Execution Environments 💼
Intel Brief: Trusted Execution Environments (TEEs)
Verida presently uses Marlin Protocol to deploy TEEs for private AI use cases.
See - Intro to Enclaves
Key Points:
TEEs provide isolated execution and memory encryption within processors
Widely implemented by major tech companies (Apple, Intel, AMD, ARM, cloud providers)
Bridge gap between blockchain limitations and need for private, complex computations
Applications in crypto:
MEV mitigation (Flashbots' SUAVE)
Order flow auctions
DePIN networks (e.g., Verida)
Wallet security (Clave, Capsule, Ledger)
zkTLS implementation (Clique)
Layer-2 solutions (Scroll, Taiko)
Challenges:
Require trust in hardware manufacturers
Vulnerability to side-channel attacks
Past security breaches (e.g., SGAxe attack on Intel SGX)
Future Outlook:
Short-term value accrual in infrastructure (Marlin, Automata, Sirrah, Lit Protocol)
Long-term potential in user-facing applications leveraging TEEs
Analysis: TEEs represent a significant advancement in secure computation, particularly valuable for crypto applications requiring privacy and off-chain processing. While not without vulnerabilities, their widespread adoption and efficiency make them a key technology to watch in the evolving crypto landscape.
Article reference - Decentralised.co - Trusted Enclaves
6. Agent Debriefing: Open Source AI Summit 🎙️
Field report from Open Source AI Summit, Singapore:
Privacy concerns dominated discussions, particularly data privacy
Audience: primarily industry insiders and developers
Key points raised:
Millions of AI agents expected in near future
These agents will require access to personal data for effectiveness
Ownership of AI agents crucial for maintaining personal independence
Critical questions: data storage, access, encryption, deletion rights
Data control seen as prerequisite for AI agent ownership
Analysis: Growing tension between AI advancement and personal data sovereignty. Industry insiders grappling with implications of widespread AI adoption. Potential shift in public discourse as privacy concerns spread beyond tech circles.
Recommendation: Monitor for emerging data ownership frameworks and AI agent governance models. Assess operations and digital identity management.
7. Agent Network - Get on the Waitlist 🌐
Welcome to the network agent. We invite you to join our Private AI community.
Verida.ai HQ is always listening and learning. Reach out to one of our many channels.
Stay vigilant, Agent. Your privacy is our mission.
Spymaster Kyra.
End Transmission.