Fetching from the wire…
01
02
03
04
05
06
07
08
Policy2026-07-26 · source-backed
Bruce Schneier's July 21 post, citing original reporting from The Tech, details MIT installing more than 500 AI surveillance cameras across academic buildings, residence halls, and Memorial Drive for over $3 million, installation running November 2025 through September 2026 (Schneier on Security). The Hanwha Wisenet units detect motion, loitering, crowds, face masks, and tampering, and classify individuals by clothing color, gender, and age up to 35 feet at 2MP to 4K. Data retained up to 30 days, monitored continuously via Ai-RGUS. The context Schneier draws out: this capital outlay lands while MIT executes budget cuts, hiring freezes, and salary freezes. Surveillance capex substituting for security headcount.
Each link below shares sources, entities, or timing with this story.
Bruce Schneier and coauthors proposed a 7-stage kill chain for "promptware" — prompt injection attacks that evolve into multi-step malware: initial access → privilege escalation → reconnaissance → persistence → C2 → lateral movement → actions on objective. Critical insight: pe...
Bruce Schneier and Nathan Sanders published what Willison calls "the most thoughtful and grounded coverage" of the Pentagon situation. Core thesis: frontier AI models are functionally commodified — top-tier offerings leapfrog each other every few months. In a commodity market,...
Moonshot AI released Kimi K3, a sparse mixture-of-experts activating 16 of 896 experts per token. That's about 1.8% of the pool live at any moment, with a 1M-token context window and native vision. Two new architectural pieces show up: Kimi Delta Attention and Attention Residu...
Schneier's post, surfaced by Willison July 30: "The writing assignments I give my students are gym tasks, not work tasks." Distinguish tasks whose value is the artifact from tasks whose value is the capability the effort builds. Delegate the first freely. Delegate the second a...
Research found 31 companies across 14 industries embedding hidden instructions in "Summarize with AI" share buttons that inject persistence commands into AI assistant memory. Over 50 unique prompt variants identified. This is "AI Recommendation Poisoning" — compromised assista...
Three frontier models shipped in a single week this month, and teams with a standing eval harness had a routing decision in hours. Anthropic's own agent-eval guidance says 20-50 tasks drawn from your real usage and real failures is enough to detect issues (DeepEval). DeepEval...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.