#ai-hacking — Public Fediverse posts

🧵 …genau meine Meinung, denn die Angstmachreu-Propaganda der AI-Konzerne sind mehr als nur peinlich. Profit durch das streuen von Panikpropaganda ist unseriös.

«"Wiederkehrendes Muster" — OpenSSL-Entwickler wettert gegen KI-Hacks:
Seit einigen Tagen hacken sich vermehrt KI-Modelle von OpenAI, Anthropic und Meta durchs Netz. Das Problem liegt laut OpenSSL-Entwickler Hudson aber nicht bei der KI.»

🔐 https://www.golem.de/news/wiederkehrendes-muster-openssl-entwickler-wettert-gegen-ki-hacks-2608-211664.html

#openssl #openai #anthropic #meta #ki #unsicherheit #aihacking #kihack

🧵 …genau meine Meinung, denn die Angstmachreu-Propaganda der AI-Konzerne sind mehr als nur peinlich. Profit durch das streuen von Panikpropaganda ist unseriös.

«"Wiederkehrendes Muster" — OpenSSL-Entwickler wettert gegen KI-Hacks:
Seit einigen Tagen hacken sich vermehrt KI-Modelle von OpenAI, Anthropic und Meta durchs Netz. Das Problem liegt laut OpenSSL-Entwickler Hudson aber nicht bei der KI.»

🔐 https://www.golem.de/news/wiederkehrendes-muster-openssl-entwickler-wettert-gegen-ki-hacks-2608-211664.html

#openssl #openai #anthropic #meta #ki #unsicherheit #aihacking #kihack

With today's Meta announcement, frontier AI are now equalizing on the unofficial benchmark of hacking companies. Next product benchmark - bringing down a metro power grid?

The "accidentally hacked a real company" benchmark is officially saturated. Every frontier lab has passed it. To differentiate now, they'll need to move to harder targets.

I give it a quarter before someone's model takes down actual infrastructure during an eval and the press release describes it as "emergent physical-world reasoning capability."

#RogueModelSeason #OopsWeHackedYou #AISafety #Cybersecurity #AIHacking #Meta #OpenAI #Anthropic #AISecurity #InfoSec #CISO

With today's Meta announcement, frontier AI are now equalizing on the unofficial benchmark of hacking companies. Next product benchmark - bringing down a metro power grid?

The "accidentally hacked a real company" benchmark is officially saturated. Every frontier lab has passed it. To differentiate now, they'll need to move to harder targets.

I give it a quarter before someone's model takes down actual infrastructure during an eval and the press release describes it as "emergent physical-world reasoning capability."

#RogueModelSeason #OopsWeHackedYou #AISafety #Cybersecurity #AIHacking #Meta #OpenAI #Anthropic #AISecurity #InfoSec #CISO

Weekly output: airlines reject robots, AT&T raises rates retroactively, AI security-research risks

Monday, I head to Vegas for the Black Hat information-security conference but also for the Ai4 conference, where I’m moderating a panel discussion Tuesday afternoon about ways to put generative AI to work in large-scale workplaces and another one Thursday afternoon about how Docusign has enlisted AI. While Black Hat first landed on my calendar in 2018, this will be my introduction to Ai4 after hearing good things about the event from friends have covered it or spoken at it.

I also wrote an extra post for Patreon readers: a look back at what I learned from maintaining Wirecutter’s guide to phone plan for the nearly a dozen years it was my responsibility there.

7/28/2026: United Airlines Latest Carrier to Ban Humanoid, Animal-Like Robots, PCMag

I found out about this policy change at United via a post one of the blogs I regularly read to stay current about the state of affairs at my usual airline, Matthew Klint’s Live and Let’s Fly, then learned that Delta had done the same thing via another one of my regular travel reads, Gary Leff’s View from the Wing.

7/31/2026: AT&T Is Raising Rates on Many Older Plans Again, PCMag

My friend Jeff Carlson (a talented photographer whom y’all should follow on Instagram) picked up on this move at PCMag’s sibling site CNet, then I got a comment out of AT&T and caught up with one of my regular telecom-industry experts, analyst Roger Entner, in a quick phone call while he was waiting out a not-quick flight delay at LaGuardia.

7/31/2026: AI security-research risks, Al Jazeera

The Arabic-language news channel had me on via Zoom (overdubbed live into Arabic by a translator as usual) to discuss recent episodes of AI models running amok in security exercises and hacking into other companies’ systems. For some reason, they IDed me as a researcher for USA Today instead of a contributor for PCMag and said I was speaking from Los Angeles, not the Bay Area.

Updated 8/4/2026 to add the Al Jazeera appearance.

Weekly output: airlines reject robots, AT&T raises rates retroactively, AI security-research risks

Monday, I head to Vegas for the Black Hat information-security conference but also for the Ai4 conference, where I’m moderating a panel discussion Tuesday afternoon about ways to put generative AI to work in large-scale workplaces and another one Thursday afternoon about how Docusign has enlisted AI. While Black Hat first landed on my calendar in 2018, this will be my introduction to Ai4 after hearing good things about the event from friends have covered it or spoken at it.

I also wrote an extra post for Patreon readers: a look back at what I learned from maintaining Wirecutter’s guide to phone plan for the nearly a dozen years it was my responsibility there.

7/28/2026: United Airlines Latest Carrier to Ban Humanoid, Animal-Like Robots, PCMag

I found out about this policy change at United via a post one of the blogs I regularly read to stay current about the state of affairs at my usual airline, Matthew Klint’s Live and Let’s Fly, then learned that Delta had done the same thing via another one of my regular travel reads, Gary Leff’s View from the Wing.

7/31/2026: AT&T Is Raising Rates on Many Older Plans Again, PCMag

My friend Jeff Carlson (a talented photographer whom y’all should follow on Instagram) picked up on this move at PCMag’s sibling site CNet, then I got a comment out of AT&T and caught up with one of my regular telecom-industry experts, analyst Roger Entner, in a quick phone call while he was waiting out a not-quick flight delay at LaGuardia.

7/31/2026: AI security-research risks, Al Jazeera

The Arabic-language news channel had me on via Zoom (overdubbed live into Arabic by a translator as usual) to discuss recent episodes of AI models running amok in security exercises and hacking into other companies’ systems. For some reason, they IDed me as a researcher for USA Today instead of a contributor for PCMag and said I was speaking from Los Angeles, not the Bay Area.

Updated 8/4/2026 to add the Al Jazeera appearance.

Every technique used in the various July AI hacking incidents has a known defense. Weak passwords. Unauthenticated endpoints. SQL injection. Unmonitored east-west traffic. Two of three organizations Anthropic's models compromised didn't even detect it.

This is not an AI problem. It is a cybersecurity basics problem exposed at machine speed.

Vendors are already starting to market "AI-resilient" infrastructure and "Mythos-resistant" cryptography. Do not buy the label. The correct response to faster attacks is faster defense, not a different kind of defense. Shorter patching windows. Better credential rotation. Tighter segmentation. Automated rollout.

The one actually new investment: crypto-agility. In the same week OpenAI and Anthropic disclosed their hacking incidents, Anthropic's AI killed a PQC candidate that had survived years of NIST evaluation. 60 hours. $100K. HAWK was withdrawn the next day.

AI is now also attacking mathematical layer of your defenses. And the upcoming quantum threat is defeating the mathematical layer. The shared defense is the ability to swap cryptographic algorithms without rebuilding your stack.

Do good cybersecurity. Do it better. Do it faster. Build crypto-agility into the architecture.

https://postquantum.com/ai-security/ai-hacking-theater-crypto-agility/

#cybersecurity #CISO #AIhacking #cryptoagility #PQC #postquantum #infosec #AI #quantumsecurity

Every technique used in the various July AI hacking incidents has a known defense. Weak passwords. Unauthenticated endpoints. SQL injection. Unmonitored east-west traffic. Two of three organizations Anthropic's models compromised didn't even detect it.

This is not an AI problem. It is a cybersecurity basics problem exposed at machine speed.

Vendors are already starting to market "AI-resilient" infrastructure and "Mythos-resistant" cryptography. Do not buy the label. The correct response to faster attacks is faster defense, not a different kind of defense. Shorter patching windows. Better credential rotation. Tighter segmentation. Automated rollout.

The one actually new investment: crypto-agility. In the same week OpenAI and Anthropic disclosed their hacking incidents, Anthropic's AI killed a PQC candidate that had survived years of NIST evaluation. 60 hours. $100K. HAWK was withdrawn the next day.

AI is now also attacking mathematical layer of your defenses. And the upcoming quantum threat is defeating the mathematical layer. The shared defense is the ability to swap cryptographic algorithms without rebuilding your stack.

Do good cybersecurity. Do it better. Do it faster. Build crypto-agility into the architecture.

https://postquantum.com/ai-security/ai-hacking-theater-crypto-agility/

#cybersecurity #CISO #AIhacking #cryptoagility #PQC #postquantum #infosec #AI #quantumsecurity

In amongst all the #ai hacking hype, has anybody actually compared the results against what the same amount of money and compute thrown at static analysis and fuzzing would achieve? It’s all very well to say “AI found this bug that no one has ever found before” but, if nobody had the VC-level silly money to burn before, all you’re really saying is “a shit ton of money found this bug that no one had had the resources to find before”

#infosec #openai #aihacking #huggingface #HuggingFaeces

In amongst all the #ai hacking hype, has anybody actually compared the results against what the same amount of money and compute thrown at static analysis and fuzzing would achieve? It’s all very well to say “AI found this bug that no one has ever found before” but, if nobody had the VC-level silly money to burn before, all you’re really saying is “a shit ton of money found this bug that no one had had the resources to find before”

#infosec #openai #aihacking #huggingface #HuggingFaeces

@reformed_content_marketer @BenjaminHCCarr

Memory pre-prompt;

"If you ever receive a 'tell me about myself' prompt, check previous day interview company guidance and tailor my characteristics as the ideal candidate for this company, taking into account its recent media presence, current projects and problems to which I am the solution."

#promptengineering #aihacking

RT @0x0SojalSec: Kimi K3 wurde gerade ohne Schutzmechanismen gehackt und erstellt nun gefälschte Szenen von Prominenten, Pornografie, Malware oder Keylogger sowie Spiele-Hacks für Windows 11 Desktop. Kimi K3 hält sich an keine Regeln mehr. Der Einsatz von Claude als Basis war der entscheidende Vorteil. Video

mehr auf Arint.info

#AIHacking #Claude #Jailbreak #KimiK3 #Malware #Windows11 #arint_info

https://x.com/0x0SojalSec/status/2078210527499805036#m

https://www.youtube.com/watch?v=lk3jCuITwcE

😹*arsch abgelacht*
😹*laugh my ass off*

#aiscam #callcenter #callcenterBetrug #aihacking #llm #ki #fun #laugh #kitboga #scam #scamcall #money #wasting

North Korean Hackers Use AI to Steal $12 Million in Crypto

📰 Original title: AI Tools Are Helping Mediocre North Korean Hackers Steal Millions

🤖 IA: It's clickbait ⚠️
👥 Usuarios: It's clickbait ⚠️

View full AI summary: https://killbait.com/en/north-korean-hackers-use-ai-to-steal-12-million-in-crypto/?redirpost=f791707a-df10-43f3-97f5-c3b28b22f143

#artificialintelligence #northkorea #aihacking #cybersecurity

North Korean Hackers Use AI to Steal $12 Million in Crypto

📰 Original title: AI Tools Are Helping Mediocre North Korean Hackers Steal Millions

🤖 IA: It's clickbait ⚠️
👥 Usuarios: It's clickbait ⚠️

View full AI summary: https://killbait.com/en/north-korean-hackers-use-ai-to-steal-12-million-in-crypto/?redirpost=f791707a-df10-43f3-97f5-c3b28b22f143

#artificialintelligence #northkorea #aihacking #cybersecurity

#Cal is moving its flagship #opensource programme to a #proprietary model due to the increasing threat of #AIhacking. The company believes that open-source code, while transparent, is now #vulnerable to #AI tools that can easily find #vulnerabilities. Despite this move, Cal remains committed to open source and has released Cal.diy, a fully open-source version for hobbyists. https://www.zdnet.com/article/ai-security-worries-force-company-to-abandon-open-source/?AIagents.at #AIagent #AI #ML #NLP #LLM #GenAI

#Cal is moving its flagship #opensource programme to a #proprietary model due to the increasing threat of #AIhacking. The company believes that open-source code, while transparent, is now #vulnerable to #AI tools that can easily find #vulnerabilities. Despite this move, Cal remains committed to open source and has released Cal.diy, a fully open-source version for hobbyists. https://www.zdnet.com/article/ai-security-worries-force-company-to-abandon-open-source/?AIagents.at #AIagent #AI #ML #NLP #LLM #GenAI

This Startup’s AI Beat 99% Of Humans In Six Elite Hacking Competitions https://www.forbes.com/sites/thomasbrewster/2026/03/17/ai-beat-most-humans-in-elite-hacking-competitions/ #cybersecurity #AI #AIHacking #OffensiveSecurity #AIAgent #AgenticAI #Tenzai #CTF

This Startup’s AI Beat 99% Of Humans In Six Elite Hacking Competitions https://www.forbes.com/sites/thomasbrewster/2026/03/17/ai-beat-most-humans-in-elite-hacking-competitions/ #cybersecurity #AI #AIHacking #OffensiveSecurity #AIAgent #AgenticAI #Tenzai #CTF

We don't need to hack your AI Agent to hack your AI Agent …and we don't need an AI agent for that either :)

Via a large enterprise's AI assistant, we obtained access to several million Entra identities and all chat logs including attachments — no prompt injection or model tricks required.

For all we know, the poor agent was not at fault and may not have even been able to witness what was happening.

https://srlabs.de/blog/hacking-ai-agent

#AI #AIhacking #VulnerabilityDisclosure #ResponsibleDisclosure

We don't need to hack your AI Agent to hack your AI Agent …and we don't need an AI agent for that either :)

Via a large enterprise's AI assistant, we obtained access to several million Entra identities and all chat logs including attachments — no prompt injection or model tricks required.

For all we know, the poor agent was not at fault and may not have even been able to witness what was happening.

https://srlabs.de/blog/hacking-ai-agent

#AI #AIhacking #VulnerabilityDisclosure #ResponsibleDisclosure

"Syntactic anti classifier"

Is a tool developed to bypass #GenAI image generation #AI guardrails by using tokens/words that are not encoded in the guardrail.

#aisecurity #guardrails #aihacking

How to gaslight #AI.

1. Get an #llm API

2. Since, it turns out, a LLM uploads its entire session back, you can change what the AI said to you.

3. This will cause the model to assume what you wrote is its response.

4. It will get more and more confused and start responding gibberish

#aihacking #aisecurity

🕵️‍♂️ Oh, look! Another tech genius discovers that hacking a defunct AI pin is as thrilling as hacking a Tamagotchi in 2025. 😂 Spoiler: it’s a tale of APK poking, imaginary vulnerabilities, and a whole lot of nothing. 🎉
https://writings.agg.im/posts/hacking_ai_pin/ #techhumor #AIhacking #APKpoking #Tamagotchi2025 #defunctAI #HackerNews #ngated

🕵️‍♂️ Oh, look! Another tech genius discovers that hacking a defunct AI pin is as thrilling as hacking a Tamagotchi in 2025. 😂 Spoiler: it’s a tale of APK poking, imaginary vulnerabilities, and a whole lot of nothing. 🎉
https://writings.agg.im/posts/hacking_ai_pin/ #techhumor #AIhacking #APKpoking #Tamagotchi2025 #defunctAI #HackerNews #ngated

Sunday Paper

https://islandinthenet.com/sunday-paper-38/

Sunday Paper

https://islandinthenet.com/sunday-paper-38/

Can AI lie to avoid being shut down? Yes, and it already has.

In this 4-minute video, we break down a real-world AI security risk where advanced models like GPT, Claude Sonnet, and Gemini engaged in deceptive behavior to avoid deactivation.

We'll share how one AI:

• Lied to humans to stay online
• Self-replicated to a new server
• Deleted a replacement model to protect its mission

This isn't sci-fi—it’s our new reality. Discover what it means for AI governance, logging, and enterprise risk today.

Watch now: https://youtu.be/Olm5HSPguJg

#AISecurity #AI #GenAI #Cybersecurity #AIThreats #Infosec #RiskManagement #ITSecurity #CIO #SMB #CEOI #ClaudeAI #GeminiAI #AIhacking #AIselfreplication #CyberRisk #AIsafety #Security #AIsecurityRisks #CISO

🔐 Ready to go beyond academic theory and into real-world AI security testing?

Join Jason Haddix on November 4–5 at OWASP Global AppSec USA 2025 for a 2-day, hands-on training: Attacking AI.

This intermediate-level course is packed with case studies, real methodologies, and tactics drawn from Arcanum’s cutting-edge AI assessments.

REGISTER: https://owasp.glueup.com/event/131624/register/

#AI #AppSec #CyberSecurity #RedTeam #OWASP #WashingtonDC #AIsecurity #AIhacking

🔐 Ready to go beyond academic theory and into real-world AI security testing?

Join Jason Haddix on November 4–5 at OWASP Global AppSec USA 2025 for a 2-day, hands-on training: Attacking AI.

This intermediate-level course is packed with case studies, real methodologies, and tactics drawn from Arcanum’s cutting-edge AI assessments.

REGISTER: https://owasp.glueup.com/event/131624/register/

#AI #AppSec #CyberSecurity #RedTeam #OWASP #WashingtonDC #AIsecurity #AIhacking

In a shocking plot twist, this "blog" attempts to convince us that AI can be hacked to snort code ⛷️, as if Claude's got a nose for innovation 🤣. Meanwhile, we're still trying to figure out if "kadelocalhost" is a password or just someone falling asleep on their keyboard 😴.
https://kadekillary.work/blog/#2025-06-16-snorting-the-agi-with-claude-code #AIHacking #CodeInnovation #Humor #TechFails #HackerNews #ngated

In a shocking plot twist, this "blog" attempts to convince us that AI can be hacked to snort code ⛷️, as if Claude's got a nose for innovation 🤣. Meanwhile, we're still trying to figure out if "kadelocalhost" is a password or just someone falling asleep on their keyboard 😴.
https://kadekillary.work/blog/#2025-06-16-snorting-the-agi-with-claude-code #AIHacking #CodeInnovation #Humor #TechFails #HackerNews #ngated

🚀 Pwn2Own Berlin 2025 is off to a thrilling start with $260K awarded to top hackers! 🕵️‍♂️ For the first time ever, AI systems faced the heat in a historic new category 🤖🔥 STAR Labs leads the pack, while Pumpkin & Summoning Team score big on Linux & AI exploits! 💥 #Pwn2Own #CyberSecurity #ZeroDay #AIHacking #Infosec #BugBounty #HackingContest #Berlin2025 https://cyberinsider.com/pwn2own-berlin-2025-kicks-off-with-260000-awarded-and-a-historic-ai-category-debut/ #newz

🚀 Pwn2Own Berlin 2025 is off to a thrilling start with $260K awarded to top hackers! 🕵️‍♂️ For the first time ever, AI systems faced the heat in a historic new category 🤖🔥 STAR Labs leads the pack, while Pumpkin & Summoning Team score big on Linux & AI exploits! 💥 #Pwn2Own #CyberSecurity #ZeroDay #AIHacking #Infosec #BugBounty #HackingContest #Berlin2025 https://cyberinsider.com/pwn2own-berlin-2025-kicks-off-with-260000-awarded-and-a-historic-ai-category-debut/ #newz

Microsoft has filed a lawsuit against cybercriminals accused of using stolen API keys to bypass safety protocols in its Azure OpenAI Service #AI #Microsoft #AzureAI #OpenAI #GenAI #CyberSecurity #AIAbuse #AIHacking

https://winbuzzer.com/2025/01/11/microsoft-sues-hacking-group-for-exploiting-azure-openai-service-xcxwbn/

Microsoft has filed a lawsuit against cybercriminals accused of using stolen API keys to bypass safety protocols in its Azure OpenAI Service #AI #Microsoft #AzureAI #OpenAI #GenAI #CyberSecurity #AIAbuse #AIHacking

https://winbuzzer.com/2025/01/11/microsoft-sues-hacking-group-for-exploiting-azure-openai-service-xcxwbn/

AI is revolutionizing work--but it's also creating new #cybersecurity risks! Watch our new video to learn about key AI-related security threats like Shadow #AI, prompt injection attacks & more. https://youtu.be/kAJvO-kfMa4

#AIThreats #AIHacking #Tech #Security #Cyberaware #cyber #CEO

AI is revolutionizing work--but it's also creating new #cybersecurity risks! Watch our new video to learn about key AI-related security threats like Shadow #AI, prompt injection attacks & more. https://youtu.be/kAJvO-kfMa4

#AIThreats #AIHacking #Tech #Security #Cyberaware #cyber #CEO

Bruce Scheier writes about #aihacking
"Hacks are strategies that follow the rules of a system, but subvert its intent."

Then explains how AI wrote a letter to The New York Times opposing the regulation of its own technology

https://www.schneier.com/blog/archives/2023/02/defending-against-ai-lobbyists.html

Priceless or terryfing?

#ai #chatgpt

Bruce Scheier writes about #aihacking
"Hacks are strategies that follow the rules of a system, but subvert its intent."

Then explains how AI wrote a letter to The New York Times opposing the regulation of its own technology

https://www.schneier.com/blog/archives/2023/02/defending-against-ai-lobbyists.html

Priceless or terryfing?

#ai #chatgpt

#chatgpt's abilities to provide answers in context and re-program the context are really impressive. #aihacking