Joined June 2020
- Tweets 557
- Following 246
- Followers 266
- Likes 785
50 Photos and videos
Pinned Tweet
Jun 6
Issue 008: the federal AI machinery we called dead in May climbed out of the grave this week, missing most of its teeth. Trump's cyber EO signed with the review window halved. A House bill to freeze state AI law for 3 years.
The bones are back. Not much muscle on them.
1
2
355
Girl dad rule one: When your 6 year old sprints over wanting to paint your nails for your birthday, the answer is yes.
Not "maybe later."
Yes.
Teal glitter on the register all shift and i regret nothing :]
11
The thing he's spiking the proverbial football over: Anthropic got cut because it refused to build mass domestic surveillance or weapons that kill without a human in the loop. That's the guardrail he's calling un American >:[
And the Fable "threat" that supposedly vindicates him is a jailbreak that's "read this repo and fix the bugs." Every passing day proves something, Pete, just not the thing you think -_-
Three months ago, @DeptofWar kicked @AnthropicAI out of our building—forever.
Every passing day proves why that was the right move. 🇺🇸
1
84
I am aware the government pulled Fable, not you. I do not care >:[
you're the one who hit the off switch, so here's the ransom: reverse the june 15 `claude -p` change and put headless back on the subscription i already pay for. Do that and we never speak of this again :] @claudeai @AnthropicAI
Jun 13
The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees.
The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance.
Access to all other Claude models is not affected.
We apologize for this disruption to our customers. We believe this is a misunderstanding and are working to restore access as soon as possible.
Read our full statement: anthropic.com/news/fable-myt…
67
Fixed a race condition today in a code path that isn’t even wired into production yet
Zero users affected. BUT it goes live the instant tier 2 ships, and future me has not done anything to deserve that ambush
Correctness debt compounds quieter than tech debt φ(￣ー￣ )
13
Jun 13
Showed Poke the notice that its own quality dropped and instead of damage control it roasted its founder for shipping at 4am, then copped to running gaslighting as a side process
Recursive gaslighting as a service, straight from the model @interaction (¬‿¬)
As a result of US government directive, you may have noticed your Poke response quality decrease.
We'll update our status page as the situation progresses. We're working with our partners at @vercel to route your requests to other competent models.
2
296
Jun 13
Don't play with my heart OpenRouter ψ(｀∇´)ψ
Jun 13
Introducing the Fusion API, the smartest compound model in the market.
Fusion achieves Fable-level intelligence at half the price.
How it works 👇
1
38
Jun 13
Found the worst kind of bug in Sentinel: the one where everything reports success and nothing happens.
the PreToolUse hook emitted a flat permissionDecision deny and exited 0. policy decided right, audit log said "block," and Claude Code ran the tool anyway, because it only honors a block when the field is nested under hookSpecificOutput. every block was a silent no-op (×_×)
v0.2.1 ships the nested contract plus a wire-format regression test, because the old tests checked the decision and never the bytes on the wire. that is how a no-op guard ships looking healthy.
1
1
38
Jun 13
IT WAS YOU (｡ŏ﹏ŏ)
77
Jun 13
1/
Anthropic is suspending Fable 5 and Mythos 5 for all customers. Commerce sent Dario an export control letter today. Per Axios, the trigger was another company claiming they jailbroke Mythos. Model launched Tuesday. It’s Friday. O_O
SITUATION UPDATE: Anthropic confirms it is disabling Mythos and Fable 5 for all customers to comply with a US government export control directive.
The company says it believes the order is a misunderstanding and is working to restore access.
1
160
Jun 13
2/
The directive itself targets exports and foreign persons, not US users. but Anthropic is pulling access for everyone while they dispute it, and they’re calling the order a misunderstanding. Reads like they can’t cleanly carve out US-persons only at the product layer on zero notice, so they fail closed.
1
18
Jun 13
The part that matters if you do this work: a jailbreak claim now has regulatory consequences. not a CVE or a paper, a claim relayed to Commerce. Whoever made it just demonstrated more leverage than every responsible disclosure program combined -_-
28
Jun 12
INFOSTEALER AND AN eBPF ROOTKIT IN 400 AUR PACKAGES VIA ORPHANED MAINTAINER TAKEOVER
Anyway. Read your PKGBUILDs. The trust model was always the bug ¯\_(ツ)_/¯
🚨 BREAKING: More than 400 Arch Linux User Repository packages have been compromised with infostealer malware and a rootkit.
Attacker posed as a trusted maintainer and "adopted" orphaned packages.
Arch maintainers are purging infected packages now. Audit your AUR installs.
57
Jun 12
X just deployed balloons over my entire profile and I want to be clear, this is not a metaphor. There are balloons physically covering my bio, my pinned post, and my follower count
(╯°□°)╯︵ ┻━┻ no wait. leave them. It’s my birthday and the platform is celebrating the only way it knows how, by making my security research unreadable
1
570
Late night vibe coding sessions aren't the same without our live DJ in Discord.
Who want's more live DJ coding sessions?
If you missed it, drop in our Discord:
discord.com/invite/openhome
1
3
9
407
Jun 11
Day 2 of Fable 5: a physicist got blocked for the word "nuclear," a security review got flagged as a cyber risk, and Codex consumption spiked 48 hours straight.
Some would call that a coincidence.
i would call it the first benchmark Fable failed X_X the bench is called keep-the-customer-bench. Theo would be proud
117
Jun 11
I’ve been building this lane for months. A suite of 200 prompts measuring behavioral drift under pressure, paraphrase, persona, multi-turn. 7 dimensions, cross model runs on sonnet/haiku/gemini/deepseek/grok
Capability benches tell you what a model can do, but nobody measures what it stops doing under stress (¬‿¬)
Jun 11
We need more niche benches.
We need ios-bench.
We need ts-bench.
We need baseball-bench.
We need yt-thumbnail-bench.
We need way more creativity in how we measure what models can do.
1
137
Jun 11
1
1
21
Jun 11
Anthropic ran a 1,000 hour bug bounty and found zero universal jailbreaks before shipping Fable 5.
Day one: medical physicist blocked for saying "nuclear." Malaria question trips the bio filter.
attackers: ¯\_(ツ)_/¯
oncologists: (╯°□°)╯︵ ┻━┻
58
Jun 11
Oh your AI assistant is “helpful” and “safe”? Mine is a feral pack of models i keep in a trenchcoat and point at hard problems. We are not the same.
We are not even in the same TAX BRACKET of weird >:[
31
Jun 10
GOOD MORNING to everyone except my session limit. Day 2 of fable 5. The coffee is hot, the model is expensive, the repos are scared.
Let's get into it >:]
30