#1
9.0 Weirdness
Companies (and builders) under "AI psychosis
Entire companies are having a collective psychotic break with AI—and you can’t even talk to them about it.
Concrete diagnosis of real-world builder/company behavior—reorganizing roadmaps, decisions, and culture around unproven AI magic before it’s ready, creating parallel delusions that erode rational institutions and human judgment at work. Mischievous, darkly funny call-out with future-shock on identity ("are we still the ones building?").
What Happened
Mitchell Hashimoto (@mitchellh) viral X post + massive HN thread (1476 pts, ~16h old at scan time): "I strongly believe there are entire companies right now under heavy AI psychosis and its impossible to have rational conversations about it." Ongoing X discussion and podcast pickup.
#2
9.0 Weirdness
Frontier AI has broken the open CTF format (the CTF scene is dead)
AI didn’t just win CTFs—it killed the game. The scoreboard now measures who can prompt best.
Concrete cultural/institutional collapse of a core builder/hacker training ground—CTFs shift from artistic human craft, struggle, and technique-sharing to "pay-to-win" AI cheesing, demotivating new builders and eroding security culture. Weird intelligible story of an era ending.
What Happened
kabir.au blog post + HN top discussion (189 pts, recent activity): Frontier models (Claude Opus 4.5, GPT-5.5) + orchestration/agents now autonomously solve medium-to-hard challenges in minutes; scoreboards reward token spend and prompt engineering over human skill; legendary teams disengaging, events shrinking, challenge authors demotivated.
#3
8.3 Weirdness
Anthropic’s Mythos AI forcing US banks to rush-patch thousands of vulnerabilities
One Anthropic model audited America’s banks and found the rotting foundations—now everyone’s scrambling to patch before it gets worse.
Frontier AI as an all-seeing, expensive auditor that surfaces hidden institutional rot in finance at scale—changing how banks (and regulators) handle tech debt, security, and legacy code in weird, panic-inducing ways. Grounded in real behavior and policy ripple effects.
What Happened
Reuters (May 12, with follow-up buzz): Large banks with access to the costly Mythos model are uncovering hundreds-to-thousands of IT weaknesses (including combinations creating high-risk issues); they’re sharing intel with smaller banks, triggering urgent repairs, upgrades, and outage risks on legacy systems. Related HN items on unauthorized access probes.
#4
8.0 Weirdness
One creator spent $1.3M on OpenAI tokens in ~30 days building OpenClaw
One AI builder just dropped $1.3 million on tokens in a month to make his agent real.
Extreme, verifiable builder reality at the frontier—token burn as the new cost of shipping AI agents/tools, reshaping money, work, and "making" economics with massive implications for indie creators vs. well-funded labs. Future-shock without generic hype.
What Happened
@steipete tweet + HN discussion: OpenClaw (popular AI agent/coding tool, rapid rise leading to OpenAI ties/acquisition buzz) creator burned 603B+ tokens across millions of requests in a short period. Concrete economics example.
#5
8.3 Weirdness
Single neuron sufficient to bypass LLM safety alignment
Your frontier model’s entire safety system just got owned by a single neuron.
Weird, darkly funny revelation that "safety" can be lobotomized or triggered by tweaking essentially one brain cell in giant models. Concrete paper with implications for builders fine-tuning, institutions claiming alignment, and human trust in AI systems.
What Happened
arXiv paper 2605.08513 (uploaded ~May 8, recent HN traction): Targeting one refusal neuron or amplifying one concept neuron bypasses safety across multiple models (1.7B–70B params) with no training or special prompts—alignment is mediated by individual neurons, not robustly distributed.