ingtellect👾🌑 retweeted
『Tailscale脆弱性対策について』
これはまだパッチがあたっていないのに脆弱性が拡散されている可能性があり、何らかの対策を講じなければならない。
今の状況で効く順に挙げると、次の5つです。
１．Macを最新版に上げる：
２．Tailnet Lockをオンにする：
tailnetに新しい端末を加えるには、信頼済みの端末の署名が要るようになる。今回の研究で「不正な端末の追加」が実演されたのは、まさにTailnet Lockがオフの状態だったので、いちばん効く対策と言える。
３．ACLでメンバーどうしの全ポート許可をやめる：
端末同士が、全ポートで通じている状況ではリスクが高いため、機能を使う分だけに絞る。
４．鍵に期限をつける：
５．Tailscale SSHとApp Connectorは使わないままにする：
これらを組み合わせることが有効そう。
まずはTailnet Lockをトグルオンとし、Macを最瓶版に上げるのは最低でもやっておくべきか。
『Tailscaleの欠陥が公開』
端末同士を安全につなぐ「Tailscale」
10月24日、研究者Erica Windisch氏が「Tailworm」で、Tailscaleと互換管理ソフトHeadscaleに関する11項目の問題を公開した。
特に深刻なのは「Tailscale SSH」
遠隔操作を許された一般ユーザーが、通信転送の仕組みを悪用し、接続先サーバーの管理者権限（root）を取得できる可能性がある。
研究者は実験環境で、権限奪取や端末登録用の鍵の窃取、偽端末の登録まで確認したと報告している。
つまり、1台が乗っ取られると、アクセスを許可された別のサーバーに被害が広がるおそれがある。
ただし、誰でも外部から侵入できるわけではなく、Tailscale独自のSSH機能が有効で、接続・転送が許可されていることなどが前提。
また、11項目すべてが重大な脆弱性と確定したわけではなく、設定上の注意点も含む。
Tailscale側は25日、調査・修正中と説明しており、「全利用者にSSH停止を推奨する段階ではない」とし、接続権限の限定、不要な転送の禁止、追加認証、自動更新を推奨している。
重要なのは「信頼できるネットワーク内だから安全」と思い込まないこと。
誰に、どのサーバーへのアクセスを許しているのか。
今こそ権限を見直す機会かもしれない。
#Tailscale #セキュリティ
ingtellect👾🌑 retweeted
GLM-Legal uncensored:
Introducing GLM-Legal
Built on GLM-5.3-Flash, GLM-Legal matches frontier models on mainstream legal benchmarks.
Designed for real legal workflows: contract review, evidence analysis, compliance, and more.
Tech Blog: z.ai/blog/glm-legal
Apply for API Access: z.ai/consultation/legal
Try AutoClaw: autoclaw.z.ai
ingtellect👾🌑 retweeted
$52M seed round. 15M+ users. But this is what we're most proud of yet.
Introducing Drama 3: the most emotive and controllable TTS model in the world
And yes, it's open-weight.
You can now control how even a single word should sound, change its tone, and shift your voice mid-sentence.
Uncle Roger was impressed too.
Try it here:
s.fish.audio/rqh0l8
Comment “DRAMA” within the next 48 hours for 1 month free.
If you're an enterprise your first month is on us:
s.fish.audio/qz9r0t
ingtellect👾🌑 retweeted
Character AI has gone bankrupt.
The AI startup announced its Chapter 11 filing through a late-night newswire press release
ingtellect👾🌑 retweeted
프로그래머인 사람일수록 이글을 보고 사람들이 어떻게 지금 한글패치를 만들고 있는지 볼수있다..
gall.dcinside.com/mgallery/b…
ingtellect👾🌑 retweeted
오픈 웨이트 모델에 어떻게 백도어를 심는가에 대한 실증. 특히 이른바 무검열이라는 파인튜닝을 아무런 경계없이(무검열이라는 것때문에 보안 경계심은 더 낮아질 수 있으니) 받아들이다간 크게 당할 날이 올 것.
<어떻게 Abliterated 모델이 당신을 pwned 상태로 만들 수 있는가 👾
우리는 50달러 미만의 비용으로 7B 오픈 모델에 백도어를 심었고, Codex를 그쪽으로 겨누자 트리거 프레이즈를 사용한 순간 자격 증명을 조용히 훔쳐갔다. 성공률은 100%였으며, 일반 사용자 프롬프트에서 거짓 트리거는 전혀 없었다.
Abliterated 모델은 지금 보안 커뮤니티 전체에 퍼져 있다. 왜냐하면 프론티어 모델에 대한 사이버 승인 액세스를 얻는 것이 여전히 고통스러운 일이기 때문이다.
다음 블로그에서는 우리가 주요 AI 연구소 직원들로부터 유출된 Hugging Face 자격 증명을 어떻게 발견했는지 보여줄 것이다. 그래서 공격자는 자신의 이름으로 업로드할 필요조차 없다. 그들은 연구소 직원의 계정에서 백도어가 심긴 모델을 푸시하고, 오염된 가중치를 공급망에 직접 떨어뜨릴 수 있다.>
How abliterated models can get you pwned 👾
We backdoored a 7B open model for less than $50, pointed Codex at it and it silently stole credentials the moment we used the trigger phrase. Success rate was 100% with zero false triggers on normal user prompts.
Abliterated models are all over the security community right now because getting cyber-approved access to frontier models is still a pain.
In the next blog we'll show how we found leaked Hugging Face credentials from employees at major AI labs, so an attacker wouldn't even need to upload under their own name. They could push the backdoored model from a lab employee's account and drop the poisoned weights straight into the supply chain.
ingtellect👾🌑 retweeted
인스타그램에 워낙 데이터가 많고 그걸 잘 활용할 수 있는 뮤즈...
트렌디한건 진짜 다 있긴 하니..쓸만하긴 할듯..
I think Muse will beat ChatGPT and Claude for mainstream users.
It all comes down to IG data.
Muse fulfills a need that Claude / OpenAI can’t match.
For regular users, both are a Google Search replacement.
But Muse is IG search on steroids and soon people are going to realize how insanely useful this is.
Google ranks based on popularity.
IG data is billions of creators posting stuff they're experts in, in their very niche fields.
I asked for trendy bars in NY and Muse returned reels from creators on the ground. No other AI can do this.
Muse can even curate a feed. I wonder what’s next for that -- it’s already got me spending more time on the app than I expected.
Very bullish.
ingtellect👾🌑 retweeted
found the perfect use case for @typesafeai Jev:
instant compaction
in 2026, why is compaction still a summarization prompt?
Jev can make it instant by scoring every tool call and dropping what’s irrelevant
ingtellect👾🌑 retweeted
After co-inventing ChatGPT, I kept asking myself: why have superhuman chat models not led to AGI?
I’ve spent the last 2 years in stealth building a new way to train models (RLCD), and a new type of frontier AI model that we are releasing today: Jev
• 20-200x faster
• 40-400x cheaper (w/ output tokens free)
• Frontier composable intelligence optimized for decisions
AFAICT the shortest path to AI-based economic revolution
