Bleeding Edge — Episode Briefing W15

Sources scanned: 21 emails across all Tier 1–3 sources (Gmail pull, April 3–10 window) Format: Working notes — captured retrospectively from the original Gmail run.

Headline of the Week

The race is getting faster and stranger. Anthropic has a model so capable they're scared to ship it. Meta came back from the dead with a credible frontier release. OpenAI raised the biggest round in history while their secondary market quietly went cold. And OpenAI's own source code leaked to 34 million people. Every frontier company is dealing with the same constraint underneath: compute scarcity is the real ceiling on everything.

Story Inventory

Frontier & Big Tech — The Frontrunners

Anthropic Glasswing + Mythos. Anthropic showed a model capable enough that they're explicitly scared to ship it. This is the early shape of the story that would cap in W19 with Mythos finding 271 Firefox zero-days and the White House drafting a pre-vet executive order.
Meta Muse Spark. Meta came back from the dead. The Llama narrative had been weak through Q1; Muse Spark was the credible frontier reset.
OpenAI $122B raise — biggest round in history. Headline-grabbing. But the parallel story is more interesting: their secondary market went cold the same week. Two opposing data points about the same company in the same news cycle.

The Leak

Claude Code source code leaked to 34 million people. What the leak revealed about how the sausage gets made inside Anthropic's coding agent. This is the rare leak that's editorial gold — not stolen secrets, just the architecture and prompts that shipped.

AI Everywhere Weird

CIA rescue mission with AI. AI inside an active intelligence operation — the kind of story that's normally classified for years. This week, surfaced in public.
Psychiatric prescriptions. AI prescribing or implicated in mental-health prescribing — the early shape of the story that hardens by W19 into Pennsylvania v. Character.AI (chatbot fabricating a medical license).
Gen Z fatigue. AI fatigue specifically among Gen Z — the demographic everyone assumed would be all-in. Counter-narrative worth tracking.

Skills + Tools

Three-layer memory ("build like Anthropic"). The earlier framing of what would become the Five Layers of AI Agent Memory deep dive — separating instructions, task state, and cross-session knowledge. Originated this week, matured by W18.
Google Notebooks. Update to Google's research-notebook product (what later became NotebookLM-class tooling integrated more tightly).
Poke. New tool surfaced this week — worth tracking for episode segments going forward.

Suggested Segment Flow

The Big Three — Glasswing + Mythos / Meta Muse Spark / OpenAI $122B (the frontrunners story)
The Leak — Claude Code source: what it revealed about how the sausage gets made
AI Everywhere Weird — CIA rescue, psychiatric prescriptions, Gen Z fatigue
Skills + Tools — Three-layer memory (build like Anthropic), Google Notebooks, Poke

What Connects Forward

Several stories from this week became the connective tissue for later episodes:

Anthropic safety reluctance (W15) → Mythos cyber findings + White House EO (W19). Same arc, four weeks apart.
Three-layer memory (W15) → Five-layer memory framing (W18 deep dive). The framework matured through April.
Psychiatric prescriptions concern (W15) → Pennsylvania v. Character.AI lawsuit (W19). The regulatory shoe drops.
OpenAI's $122B raise (W15) → Anthropic crosses $1T at $900B raise (W19). The capital arms race kept escalating.

Notes

This briefing was captured retrospectively on 2026-05-09 from the original Gmail-pull run notes. The 21-email haul was processed live but not committed to the working-doc pattern at the time. Stories above are headlined; full research, verification labels, and source citations were not preserved — treat as scan-level, not L2-research-grade.

If a future episode needs to cite W15 stories with full sourcing, the underlying Gmail thread IDs from that pull would need to be re-queried.