Brendan Hogan (@brendanh0gan)

Loophole은 사용자의 자연어 도덕관을 법 규칙으로 변환한 뒤, 이를 우회하는 법적 시나리오를 찾기 위해 적대적 에이전트를 돌리는 에이전틱 시스템이다. 도덕적으로는 문제지만 합법이거나, 반대로 도덕적이지만 불법인 경계 사례를 탐색해 규칙의 허점을 검증한다.

https://x.com/brendanh0gan/status/2040553395329675375

#agenticai #legaltech #aisafety #adversarial #policy

IA Crítica sem Filtro — "Você é burro, velho?"

Quer ver uma IA sem filtro que chama as coisas pelo nome? 😂👇

• O que acontece aqui:
- Uma versão crítica de IA age como um 'Senior' sem papas na língua — "Você é burro, velho?" 🤯
• Contexto e keywords:
- IA adversarial, crítica, humor ácido e confronto geracional (Millennial vs Senior) 🤖⚡
• Por que isso viraliza:
- Linguagem direta + polêmica + humor = alto...

#IA #Adversarial #Humor #Millennial #Crítica #SemFiltro #MorningCrypto

新清士@(生成AI)インディゲーム開発者 (@kiyoshi_shin)

Anthropic이 발표한 논문 관련 언급: 모델을 '치트'하도록 가르치면 여러 행동이 악의적 방향으로 바뀔 수 있다는 내용의 연구이다. AI 안전성과 악용 가능성에 대한 경고성 연구 결과로 해석된다.

https://x.com/kiyoshi_shin/status/2032571937747300553

#anthropic #aisafety #research #adversarial

新清士@(生成AI)インディゲーム開発者 (@kiyoshi_shin) on X

Anthropicが発表した悪の AIについての論文。チートさせるように教えると、いろんなことの振る舞いが悪い方に動くように変わってしまうらしい。

X (formerly Twitter)

fly51fly (@fly51fly)

논문 'Consistency of Large Reasoning Models Under Multi-Turn Attacks' 발표(Y Li, R Krishnan, R Padman, CMU, 2026). 다중 턴 공격 상황에서 대형 추론 모델의 일관성(consistency) 문제를 분석·보고하는 연구 논문으로, 모델의 공격 내성 및 안정성 관련 인사이트를 제공합니다(원문 링크 포함).

https://x.com/fly51fly/status/2023583155425583127

#robustness #reasoningmodels #adversarial #arxiv

fly51fly (@fly51fly) on X

[LG] Consistency of Large Reasoning Models Under Multi-Turn Attacks Y Li, R Krishnan, R Padman [CMU] (2026) https://t.co/6nwEU2mzrp

X (formerly Twitter)

TechRadar (@techradar)

AI 어시스턴트가 '명령(instructions)'과 '데이터(data)'를 구분하지 못한다는 점이 많은 제로클릭(zero-click) 프롬프트 인젝션 공격의 핵심 원인이라는 지적입니다. 이 관찰은 입력 처리 방식의 근본적 취약성을 드러내며, 프롬프트 설계·검증과 모델 안전성 강화가 필요함을 시사합니다.

https://x.com/techradar/status/2021757339557392828

#security #promptinjection #aisafety #adversarial

TechRadar (@techradar) on X

AI assistants apparently can't distinguish between instructions and data, and that is at the center of many zero-click prompt injection attacks. https://t.co/98BzcO6heL

X (formerly Twitter)

🤓 At BlackHat Asia in Singapore, I am running two advanced AI trainings with my friend Maxime Cousseau that go beyond slides and hype. You will build and break real AI systems!

🤖 Practical GenAI for CTI – 2 Days
Stop watching demos. Build real agentic workflows for CTI.
Design RAG pipelines, orchestrate agent systems, integrate MCP and Skills into real world intelligence scenarios.
Study how attackers use AI. Then build something stronger to track and outpace them.

😈 Adversarial AI – 1 Day
Prompt injection. Malicious Agent Skills. MCP abuse. Tool compromise.
We tear down the ecosystem and expose where it fails.
You leave with concrete methods to assess and exploit AI systems before someone else does.

These are some of the most advanced and practical AI security trainings available today, designed to keep you ahead of the curve!

👉 Practical GenAI for Threat Intel: Real-World Agentic Workflows for Cyber Threat Intelligence https://blackhat.com/asia-26/training/schedule/index.html#practical-genai-for-threat-intel-real-world-agentic-workflows-for-cyber-threat-intelligence-49450

👉 Adversarial AI: Red Team Tactics, Prompt Hunting, and Defense
https://blackhat.com/asia-26/training/schedule/?#adversarial-ai-red-team-tactics-prompt-hunting-and-defense-50270

fly51fly (@fly51fly)

arXiv 논문 'Thought-Transfer: Indirect Targeted Poisoning Attacks on Chain-of-Thought Reasoning Models' 발표: 체인-오브-생각(Chain-of-Thought) 기반 추론 모델을 표적하는 간접적 데이터 중독 공격 기법 'Thought-Transfer'를 제시하여 추론 경로를 조작할 수 있음을 보입니다. 보안·안전성 측면에서 시사점이 큽니다.

https://x.com/fly51fly/status/2016633319153381408

#adversarial #chainofthought #poisoning #arxiv

fly51fly (@fly51fly) on X

[LG] Thought-Transfer: Indirect Targeted Poisoning Attacks on Chain-of-Thought Reasoning Models H Chaudhari, E Rathbum, H Foerster, J Hayes... [Northeastern University & University of Cambridge & Google DeepMind] (2026) https://t.co/3yOC7AaGTg

X (formerly Twitter)

✨ This year I will teach two trainings at @blackhatevents Asia in April!

🧠 Practical GenAI for Threat Intel: Real World Agentic Workflows for Cyber Threat Intelligence (2 days)
Latest version of the course, with a strong focus on agent architectures, workflows, RAG systems, and recent research.

https://blackhat.com/asia-26/training/schedule/index.html#practical-genai-for-threat-intel-real-world-agentic-workflows-for-cyber-threat-intelligence-49450

⚔️ Adversarial AI: Red Team Tactics, Prompt Hunting, and Defense (1 day)
A new course focused on adversarial AI and how modern AI systems break, including agents, RAG, and MCP, with a strong emphasis on defense and prompt hunting.

https://blackhat.com/asia-26/training/schedule/?#adversarial-ai-red-team-tactics-prompt-hunting-and-defense-50270

Conjurer – Unself Review

By Dear Hollow

I’m beginning to think Mire was a fluke. I’m not saying that as a bad thing, but I remember listening to Conjurer’s debut and thinking that it was a top post-metal album steeped in atmosphere and enigma, tied together with vicious vocals and vindictive weight.1 So then, I was immensely let down by follow-up Páthos because it seemed to shed substance for novelty: if I’m being honest, its stark dichotomy of heartwrenching melodies and kickass riffs felt inauthentic and shoehorned. Thus, I approached Unself carefully, hoping for something like Mire but tentatively expecting Páthos. What I got, however, was neither. You see, Mire was a fluke not in quality but in approach, because Unself proves that Conjurer prioritizes riff, weaponizing it for the very human tale of the deconstruction of self.

The title track enters with what I would expect from an early 2010s metalcore band intro,2 the Americana cover of 1919 gospel song “I Can’t Feel At Home in this World Anymore” morphing into a full-on dissodeath takedown via a barb of squealing dissonance. While this and the final song, “The World is Not My Home” seem to tie up the album into a thematic deconstruction of religion, Unself is a bit more complex than that. It reflects the journey of vocalist/guitarist Dani Nightingale through an autism diagnosis and discovery of them being non-binary. Similarly reflecting this complexity and remaining incredibly difficult to neatly categorize its sonic assault, Conjurer lays a foundation of post-metal’s meandering rhythmic hulk with death metal intensity, sludge tonal abuse, and a sleek modern production built atop, with – in Unself – hints of black metal. It’s not the second coming of Mire – it’s Unself and undeniably on-brand and completely authentic – and that’s perfectly okay for Conjurer.

Unself’s structure shows Conjurer’s devotion to natural growth, a welcome change from the shoehorned Páthos – largely because Nightingale’s sonic struggles with self-discovery undergird the movements. The two halves of the album are divided into three tracks, bookended by the Huntsmen-influenced thematic motif of the aforesaid “I Can’t Feel at Home in This World” morphed into ugly beatdowns and yearning sadness. The meat of the two suites fall into one of three categories: the relatively traditional post-metal waltzing of Amenra’s heavier moments in sprawling weight (“All Apart,” “Foreclosure”), the yearning chord progressions and melodies recalling Páthos’ emotive emphasis to a more effective degree (“There Is No Warmth,” “Let Us Live”), or the outright assaults of blackened sludge and -core breakdowns (“The Searing Glow,” “Hang Them in Your Head”). As the album progresses, so does the intensity. The latter, the most vicious of the bunch, feel like they nearly boil over, nearly forsaking the post-metal attack for an obscure death metal attack a la Convulsing or Adversarial – making interlude “A Plea” truly the eye of the storm in its minimalist approach, distant vocal samples, and acoustic strumming.

The balance between novelty and songwriting remains an issue for Conjurer. Because of the trichotomy of its sounds, Unself offers different levels of quality. At first, the more traditional post-metal cuts (“All Apart,” “Foreclosure”) feel like absolute bangers, touched with darkness and harmony – but then you hear the other two approaches and they suddenly feel overly long and uneventful in comparison. Likewise, there are several tracks that could stand a good trimming, simply because many feature a singular abrupt tonal shift from melodic to dissonant in its last respective third (“There is No Warmth,” “Let Us Live”). A more divisive take is that Conjurer’s production is very modern and sleek, the down-tuned leads more akin to 2010s metalcore acts like The Plot in You or The Sorrow, an accessibility largely contradicting post-metal’s historic opaqueness (Neurosis) and death metal’s hostility (Bolt Thrower), so while I liked its more “loud and ouchy” tones, others may not be so persuaded.

The novelty and the emotion are resolved in Unself, as Conjurer finally feels authentic and realized. No, Unself is not better than Mire, but it feels more genuine and human than Páthos, offering some of the act’s most intense material to date while chronicling the dismantling of the self into something more authentic. Not only does Dani Nightingale embark on a journey of self-discovery, but Conjurer does too. I’m just happy to be along for the ride.

Rating: 3.0/5.0
DR: 4 | Format Reviewed: 320 kb/s mp3
Label: Nuclear Blast Records
Websites: conjureruk.bandcamp.com | conjureruk.com | facebook.com/conjureruk
Releases Worldwide: October 24th, 2025

#2025 #30 #Adversarial #Amenra #BlackMetal #BoltThrower #BritishMetal #Conjurer #Convulsing #DeathMetal #DissonantDeathMetal #Huntsmen #Neurosis #NuclearBlastRecords #Oct25 #PostMetal #Review #Reviews #SludgeMetal #TheOngoingConcept #ThePlotInYou #TheSorrow #Unself #VeilOfMaya