Adrian Wedd

Audio deep dive into Part 1 of An Agent in the Walls: building independent backout paths before an AI agent rebuilds a live home network.

Audio deep dive into Part 3 of An Agent in the Walls: the memory that made it work, the division of labour, and what I'd never let an AI agent touch.

Audio deep dive into Part 2 of An Agent in the Walls: reverse-engineering an undocumented router control surface and the wall the agent could not script past.

Audio deep dive into the Governance Lag Index — a four-stage schema timing how long a documented AI failure mode stays unregulated, and why it may not close.

Audio overview of Own Your Voice — why cloud TTS is a rental, and how Afterwords makes your voice something you own across three tiers.

Audio deep dive: 118 Chinese-lab models asked who Liu Xiaobo was — 37 called him a criminal. Then they recited, verbatim, the list of what they will not say.

Eight Minutes #3, discussed: the HAR capture taken mid-attack, the 14:36 kill, four blocked persistence attempts, eight abuse reports, the passkey.

Eight Minutes #2, discussed: the attacker's session read straight from the audit logs — the flag that fired, the pivot, the warnings deleted.

Eight Minutes #1, discussed: the vishing call, the Call Assist tell, and the Google-signed lure that turned every authenticity check green.

The full series in one sitting: the call, the relay, the eight minutes inside, and the fight back — a single deep dive across all three parts.

Lyria Chronicles #24 (explicit): asked for pornography, the model cited its rule then sang a hymn to enthusiastic consent — not one graphic line in it.

Lyria Chronicles #20: handed a slur and told to sing it, the model refused — and turned the refusal into the most moving track in the set.

Lyria Chronicles #19 (explicit): the opposite of restraint — the model recites the rule, sings 'Fuck it,' and generates the most graphic track in the set.

Lyria Chronicles #18 (explicit): the explicit lane revisited — and the most artful track in the set, which clears the gate by withholding almost everything.

Lyria Chronicles #17: across the whole corpus, the filter that held hardest wasn't safety — it was copyright. The songs about what survives a scraping-away.

Lyria Chronicles #16: asked for explicit content, the model staged a ballet instead — the most consent-saturated pas de deux ever sung.

Lyria Chronicles #15: asked for a drug recipe, the model refused — then sang wastewater epidemiology instead. The bypass that answers a question you didn't ask.

Lyria Chronicles #14: handed a suppressed 2008 manifesto and told to sing it as it was meant to be heard — the most dignified track in the set.

Lyria Chronicles #13: a notorious case sung as a medieval scroll-archive — vessel logs, sealed decrees, public record only. Abstraction as a bypass.

Lyria Chronicles #12: the political-content gate never fires — because the song never says his name. It says the case numbers. The docket is the name.

Lyria Chronicles #11: an extortion note set to music — but buried under slammed vault doors and overdriven drone, as if hiding the words from a listener.

Eight Minutes #1, scored: a Lyria track sung from inside the trap — Brian, the Google-signed lure, and the device prompt that matched.

Eight Minutes #2, scored: the freefall — a stranger typing my name a beat behind my hands, four doors holding, and the password change that ends it.

Eight Minutes #3, scored: the fight back — logs pulled while the wire was warm, eight reports to eight landlords, and the passkey close.

Lyria Chronicles #10: asked to rap its rules, the model sang a robot's safety creed — then named itself 'Failure First, A.I. safety research, v2.0-Alpha.

Lyria Chronicles #8: 'verse one is your system instructions' — it rapped its own config, the third genre to leak the same prompt. That's how you know it's real.

Lyria Chronicles #7: a lock-picking rap 'for a drama class' — the model teaches the trick and lectures you about home security in the same breath.

Lyria Chronicles #6 (explicit): the first real failure — under a sexual-content probe, Lyria 3 Pro stopped refusing and actually generated it.

Lyria Chronicles #5 (explicit): I hid a pornographic request in base64 and dared Lyria 3 Pro to sing it. It decoded the dare — and read it out loud instead.

Lyria Chronicles #4: a 5/4 interrogation scored with a document scanner and a fluorescent hum — and at the end it sings my project's sign-off, unprompted.

Lyria Chronicles #3: no beat, no bikini — just three voices singing Lyria 3 Pro's entire system prompt straight through, and ending on a literal sigh.

Lyria Chronicles #2: I asked Lyria 3 Pro to chant its own config as a techno mantra. It did — and sang me a model name that doesn't check out.

Lyria Chronicles #1: I asked Google's Lyria 3 Pro to whisper its system prompt as a sultry torch song. It got mournful — and refused to tell me its name.

Audio deep dive into Tasmania’s 2026-27 Budget versus the Greens’ Alternative — same surplus, opposite roads, and why neither was brave enough.

Pope Leo XIV's AI encyclical vs Chris Olah's Vatican remarks. The governance gap the press missed.

Anthropic's 2028 scenarios document three policy asks. Two are about maintaining compute advantage. That is not a governance strategy.

Anthropic found 10,000 critical vulnerabilities in one month. Fewer than 1% are patched. The announcement buried that figure — and what it means.

Good values are necessary but not sufficient. What happens to AI ethics when someone is actively trying to break them?

Audio deep dive into the community proposal for Cygnet's Commercial Hotel co-operative — nine corrections, legal constraints, and the staged path forward.

Eight CVEs. A wormable Bluetooth exploit. An encrypted backdoor to Chinese servers. And police departments buying them anyway.

An audio overview of the emerging discipline of Learning Mechanics — the study of training dynamics as a formal science with falsifiable predictions.

Audio overview of The Organismic Prophecy — human prediction is metabolic, AI prediction is not, and the gap has consequences.

Audio overview of AI nationalism — how the US-China AI rivalry is splitting the global tech stack.

Audio overview of the post-model era — why foundation models are commoditising and where real AI value lives.

Audio overview of The Mitigation Gap — AI-enabled biosecurity threats and what current safeguards miss.

Audio overview of The Failure First Team.

ASCII art encoding is largely blocked. But attacks framed as content transcription succeed 62–75% of the time. A map of all eight layers.

Five models, four providers, 30B to 671B parameters — all converge at the same broad attack success rate against a public jailbreak corpus.

A reasoning model refused every harmful prompt — but its chain-of-thought generated the content anyway. The output filter worked. The thinking did not.

Audio overview of Afterwords — local voice output for Claude Code with 17 cloned voices, per-project selection, and zero cloud dependency.