EP 265.5 The Deep Dive. The IT Privacy and Security Weekly Update for November 4th., 2025 and the Purported Porch Pirate

AI agents are exploding in power and reach, simultaneously automating code security (OpenAI Aardvark), bypassing paywalls, and triggering corporate warfare (Amazon vs. Perplexity). Yet automated surveillance is failing citizens: a Colorado woman was falsely accused of theft byFlock cameras, only cleared by her Rivian’s own footage. Norway disabled internet on 850 Chinese buses after finding hidden remote-shutdown features, while Xi Jinping joked about “backdoors” when gifting Xiaomi phones to South Korea’s president—amid live U.S.-China trade tensions.

1. AI Agents & Browsers

• Atlas (OpenAI) collects every click to train models; users are the product.

• Comet (Perplexity) bypasses paywalls, slashing publisher referrals 96%; Amazon calls it fraud for undisclosed AI purchases.

• AI browsers remain clunky and vulnerable to prompt-injection attacks.

2. Autonomous Cyber Defense

• Aardvark (GPT-5) scans repos, validates exploits in sandboxes, and auto-patches; 92% detection, 10+ CVEs found.

• Edge & Chrome use on-device AI to block scareware pop-ups—no cloud, no privacy leak.

• GitHub Octoverse 2026 Forecast: AI writes >30% of code; TypeScript + Python >50% of new repos; India overtakes U.S. as #1 contributor.

3. Geopolitical Tech Risks

• Norway: 850 Chinese e-buses lose web access after remote-disable code discovered in diagnostics.

• Xi-Lee Summit: Xiaomi phone gift → “Check for backdoors” quip → laughter, but U.S. espionage fears linger.

4. Surveillance Backfire

• Colorado: Flock ALPR logs Rivian passing → police issue summons without checking timestamps.

• Rivian’s 360° cameras prove owner never stopped; charges dropped.

• Lesson: automated data treated as fact, not evidence, until countered by personal tech.

Bottom Line

AI is now infrastructure—writing code, reading paywalls, and defending systems—yet it amplifies surveillance errors and geopolitical fault lines. Tools built for control can misidentify citizens or disable cities. The same camera that accuses can exonerate; the same agent that shops can defraud. Human oversight remains the final firewall.