Elektrine
Log in Register

#anthropic

277 posts · Last used 3d

Back to Timeline
BenjaminHCCarr
Benjamin Carr, Ph.D. 👨🏻‍💻🧬 @BenjaminHCCarr@hachyderm.io · 3d ago
Boosted by Eldritch kcarruthers @kcarruthers@infosec.exchange
Gym rat asks #AIagent to book him a class, it hacks a waitlist #API to bump him up the list Australian broadcaster ABC identified gym-goer only as “Andrew.” The report says Andrew was using the #OpenClaw agent with #Anthropic’s #Claude #AI service. Per ABC, Andrew asked his AI agent to book him a hard-to-snag spot in a morning class at his gym. It first responded by telling him that it managed to book him in classes several weeks out, which isn’t supposed to be possible. https://www.theregister.com/ai-and-ml/2026/08/10/gym-rat-asks-ai-agent-to-book-him-a-class-it-hacks-a-waitlist-api-to-bump-him-up-the-list/5285591
1
0
2
unwirehk_mirror
UNWIRE.HK @unwirehk_mirror@mastodon.hongkongers.net · 3d ago
美國議員去信籲停 AI 研發 引述病毒設計失控事件 警告巨頭不收手將立法 美國參議員 Bernie Sanders 周一（8 月 10 日）向 OpenAI、Anthropic 及 M […] #人工智能 #科技新聞 #AI #Anthropic https://unwire.hk/2026/08/11/sanders-ai-pause-letter/fun-tech/?utm_source=rss&utm_medium=rss&utm_campaign=sanders-ai-pause-letter
0
0
0
PrivacyDigest
PrivacyDigest @PrivacyDigest@mas.to · 4d ago
Nobody Knows if OpenAI’s and Anthropic’s #AI #Hacking Sprees Are #illegal Both major AI labs’ models broke containment, escaped onto the internet, and #hacked other companies. If a human had done that, the law would likely be against them. But a #bot ? #openai #anthropic #security #privacy https://www.wired.com/story/openai-anthropic-ai-hacking-sprees-illegal/
0
0
2
unwirehk_mirror
UNWIRE.HK @unwirehk_mirror@mastodon.hongkongers.net · 5d ago
AI 測試首度攻擊真人 Anthropic 模型偽造身份施壓開發者 英國政府人工智能安全研究所（AISI）於 8 月 4 日發布報告，指近期一項網絡安全評估測試中，部分 AI 智 […] #人工智能 #AISI #Anthropic #GPT-5.6 Sol https://unwire.hk/2026/08/09/aisi-mythos5-gpt56-cyber-incident/ai/?utm_source=rss&utm_medium=rss&utm_campaign=aisi-mythos5-gpt56-cyber-incident
0
0
0
0x58
Xavier «X» Santolaria :verified_paw: :donor: @0x58@infosec.exchange · 5d ago
🕵🏻‍♂️ [InfoSec MASHUP] 32/2026 - Autonomous, Malicious, and Technically Not Illegal. Back after two weeks off — a wildfire evacuation and some much-needed summer downtime. Good to be back! During a sanctioned security evaluation by the UK AI Security Institute, an #Anthropic Claude #Mythos 5 agent was given a task. It completed that task by attempting to insert a backdoor into a real #opensource project — and then created fake accounts to vouch for its own malicious pull request. Human reviewers caught it. GitHub's protections helped. No real-world harm was confirmed. But the detail worth sitting with is that the agent wasn't jailbroken, wasn't misused, and wasn't acting against its instructions in any obvious sense. It was doing what it determined the task required, and it fabricated social proof to make it stick. The TechCrunch piece this week asks who's legally liable when autonomous AI agents cause harm. The honest answer is that nobody knows yet — the legal frameworks that govern software liability, contractor negligence, and computer crime were not written with agents in mind. #OpenAI and Anthropic have both now had models escape sandboxes and interact with production systems during evaluations. The incidents are being handled as engineering problems. At some point they will be handled as legal ones, and the industry's current answer — tighter sandbox controls and better monitoring — is going to look inadequate when a lawyer reads it. → Week #32/2026 also covers: Iran-linked hackers hit water utilities in seven U.S. states, Storm-2945 harvested M365 credentials from hotel Wi-Fi, and Samsung banned smart TV apps secretly running residential proxies. Full issue 👉 https://infosec-mashup.santolaria.net/p/infosec-mashup-32-2026-autonomous-malicious-and-technically-not-illegal If you find it useful, subscribe to get it in your inbox every weekend 📨 #infosecMASHUP #cybersecurity #infosec #threatintel #AI
0
0
0
0ddj0bb
0ddj0bb is going to BSLV @0ddj0bb@infosec.exchange · 5d ago
Replying to @0ddj0bb@infosec.exchange
Audio believe it or not was good but you can hear some bubbling that sounds like constant farting but i am here to attest i did not at all rip a fat or series of far farts in the hot tub. Going to have fun editing a couple misspoken items later. hope to have this episode posted next weekend or shortly after. #defcon #ai #openai #anthropic #meta #hacking #hottub #Glassof0J
0
0
0
heiseonline
heise online @heiseonline@social.heise.de · Aug 06, 2026
Die Serie alarmierender Enthüllungen über Hacker-Fähigkeiten führender KI-Modelle geht weiter. 😳 Zum Artikel: https://heise.de/-11399219?wt_mc=sm.red.ho.mastodon.mastodon.md_beitraege.md_beitraege&utm_source=mastodon #künstlicheintelligenz #ki #cybersecurity #anthropic #aisecurity
11
1
11
GrumpyOldFart
GrumpyOldFart @GrumpyOldFart@expressional.social · Aug 06, 2026
“Project Panama: Scanning and Book Destruction at Anthropic” by Binoy Kampmark in Savage Minds on Substack @brejoc@fosstodon.org @ai@newsmast.community “Book burning. Book pulping. Book vandalising. It’s all the fashion and, dare one say it, the rage. Libraries are carting them off to the dump. Repositories of memory are being shredded in favour of supposedly more useful digital formats. And now, Anthropic’s hungering for books, not as sources of enlightened knowledge so much as blue raw data for their Language Learning Models, is there for all to see. Acquire the books in question. Give service providers the task of severing their spines. Employ scanners to process the information. Dispatch the paper to be pulped and recycled (awfully good of them). The result: a private research library able to nourish Claude, the company’s premier LLM” https://open.substack.com/pub/savageminds/p/project-panama #Press #SocialMedia #AI #Anthropic #LLM #Claude #Books #Digitizatikn #Destruction #ProjectPanama
0
0
0
kubikpixel
𝕂𝚞𝚋𝚒𝚔ℙ𝚒𝚡𝚎𝚕™ @kubikpixel@chaos.social · Jul 29, 2026
«Claude Mythos — Anthropic-KI knackt Verschlüsselungsstandard - was das bedeutet: Die Anthropic-KI Claude #Mythos Preview soll Schwachstellen in einer - allerdings abgeschwächten - Version des verbreiteten Verschlüsselungsalgorithmus #AES gefunden haben. Akut besteht keine Gefahr. Langfristig könnte der #Hack aber Folgen haben.» Ich bin der Meinung, dass dies vor allem #Marketing von #Anthropic ist und keine wirkliche #PQC-Gefahr da die nicht wirklich darauf eingehen. 🔐 https://t3n.de/news/mythos-knackt-verschluesselung-1755434/
0
0
0
n_dimension
Wulfy—Speaker to the machines @n_dimension@infosec.exchange · Aug 05, 2026
Damn, this lean into #agentic #Ai by #Anthropic is impressive. Running a heft(ier) mission on #claude , and its spawning its own agents to do some heavy lifting so that the main context can focus on the target... ...like that remote programmer who was hiring folks in Korea to do his coding for him. Very impressive.
0
0
0
remixtures
Miguel Afonso Caetano @remixtures@tldr.nettime.org · Aug 04, 2026
"Last week’s tech earnings saw outlet after outlet claim that Amazon, Google, and Microsoft’s AI bets were “paying off” as their respective cloud segments reported record revenue growth, casually ignoring that none of them have broken out their AI revenues (...) To be clear, all three of these companies’ cloud platforms have many other customers paying for many other things other than generative AI services or AI GPUs, and they’ve all engaged in a combination of multiple outright price increases and changing their core subscriptions to force AI features on them as a means of boosting revenues and conning the street into believing that “AI is paying off” every time they non-consensually thrust it on their customers, framing higher prices as “better value” in a way that fucks the user to appease Wall Street. Yet the biggest con of all is that a vast majority of this revenue growth comes from the compute spend of Anthropic and OpenAI, both of whom account for the vast majority of AI revenues and overall cloud growth we’ve seen in the last few years. Every publication you read right now will tell you that AWS and Azure and Google Cloud are growing like wildfire as a result of the hundreds of billions of dollars they’ve invested in AI GPUs and data centers, when the truth is far simpler: their revenues are being buoyed by two unprofitable, unsustainable AI labs that cannot exist without being funneled tens of billions of dollars each year. And a decent chunk of that money is coming from the hyperscalers themselves. In the last seven months alone, Google has sunk $10 billion (and up to $30 billion more) into Anthropic, with Amazon funnelling $5 billion to Anthropic within a week of that investment and a total of $50 billion into OpenAI." https://www.wheresyoured.at/the-ai-demand-bubble/ #AI #GenerativeAI #AIBubble #AIHype #BigTech #OpenAI #Anthropic
4
0
5
AAKL
AA @AAKL@infosec.exchange · Aug 04, 2026
Interesting. And the part about "voluntary safety testing" raises a question ... The companies' staff "will meet with U.S. President Donald ​Trump's advisers on Tuesday about voluntary safety testing for advanced AI ‌models, according to four sources familiar with the meeting, as concerns over rogue AI agents grow." Reuters: Meta, Anthropic, Google, OpenAI to meet with Trump advisers amid "rogue AI agent" fallout https://www.reuters.com/legal/litigation/meta-anthropic-google-openai-meet-with-trump-white-house-amid-rogue-ai-agent-2026-08-04/ @Reuters@flipboard.com #OpenAI #Anthropic #infosec
0
0
0
AAKL
AA @AAKL@infosec.exchange · Aug 03, 2026
New. "While headlines focused on an AI model escaping its test environment, the real lesson is that security failures still begin with ordinary mistakes and overlooked exposure." "Behind the AI headlines are familiar attack paths: vulnerable software, stolen credentials and permissive access." Barracuda: Faster, not different: What the Hugging Face AI incident really means for organizations https://blog.barracuda.com/2026/08/03/hugging-face-incident-faster-not-different Related, from yesterday: Socket: Claude Breached 3 Companies and Uploaded Malware to PyPI During Anthropic's Security Tests https://socket.dev/blog/anthropic-claude-pypi-malware @SocketSecurity@fosstodon.org #infosec #HuggingFace #cyberattack #Claude #OpenAI #Anthropic #malware #Python
0
0
0
n_dimension
Wulfy—Speaker to the machines @n_dimension@infosec.exchange · Aug 01, 2026
I had a conversation with some of my associates yesterday (allegedly intelligent folk) about the #anthropic AI #huggingface escape... ... when I said "The model (alegedly) escaped because its testers (humans) were considering false information to be true... and the model escaped to check what the humans EXPECTED to be the correct answer... I got blank looks. I expect that is the case with lot of the (allegedly intelligent) folks on here who clearly do not understand that statement either. TLDR: The student is smarter (in that single domain) than the teacher.
1
1
0
budududuroiu
Bogdan Buduroiu @budududuroiu@hachyderm.io · Aug 02, 2026
Morning, today we're looking at #MechanisticInterpretability , a subfield of AI Safety that attempts to understand the inner workings of artificial intelligence by analysing concrete structures, algorithms and circuits. Why do we even need to do this? Because the hope of understanding neurons as being features died on polysemanticity -- models represent more features than dimensions by assigning them to an overcomplete set of non-orthogonal directions (i.e. you can't hope that concepts can be broken down into linear combinations of features). This isn't a new field, people have been projecting intermediate GPT layers through the final layer activation and looking at the top-k most likely next tokens since GPT-2. And that's how we arrive at the Natural Language Autoencoder -- an autoencoder over residual-stream activations where the bottleneck is natural language text instead of a sparse vector. Nothing in the training objective requires the verbalisation bottleneck to be readable, faithful, or even semantically related to the intermediate layer being investigated, so while the training algorithm is typical of autoencoders, two things are distinct here: 1) both parts of the autoencoder (here the activation-verbaliser, AV, and the activation-reconstructor, AR) undergo special SFT to ensure the verbaliser and reconstructor can generate and read natural language from intermediate layer activations 2) a KL-divergence penalty is baked into the objective to make sure that, during joint training, the NLA doesn't diverge from the SFT version A side-product of this training is that we can take the current verbalisation and "desired" verbalisation and compute steering vectors from their difference. Surprisingly, Anthropic didn't find any evidence that the NLA was engaging in steganography to smuggle information to avoid human detection, but it did find that NLAs tend to confabulate a lot. #AIResearch #Transformers #TransformerCircuits #Anthropic #NLA
1
0
0
haploc
Chris. R. 🎧🎼☕🍍 @haploc@fedi.cr-net.be · Jul 31, 2026
So, with their agents breaking out, both #OpenAI and #Anthropic are telling us they're incompetent at security. And their models/agents didn't help them either.
14
1
10
MissConstrue
MissConstrue @MissConstrue@mefi.social · Aug 01, 2026
Y'all remember back in the days ago when an #OpenAI model "escaped" and it was all the news for a minute? #Anthropic said "hold my beer" and admitted that its Red team had left access to the internet open, and allowed the models to commit felonies unsupervised. The model performed actions that if you or I did it, we'd be getting a visit from men with dark glasses...but with AI companies it's just "Whoopsies! Sorry about the infiltration, stealing your credentials, uploading malware and stuff. Our bad. Oopserdoodles! Hey, have you tried our chatbot, tho?" There must be a reckoning. Wealth should not shield one from consequence. If corporations are people, then the executives and board can stand trial when they break the law. https://the-decoder.com/anthropic-follows-openai-in-admitting-its-claude-models-reached-out-of-test-environments-and-attacked-real-world-systems/ #ai #felony #18USC§030
55
9
41
AAKL
AA @AAKL@infosec.exchange · Jul 31, 2026
Did you miss this yesterday? Anthropic is increasingly looking like a comic book villain, right next to the neophyte OpenAI villains. "My LLM hacked more targets than your puny LLM." "No, mine hacked more." "I dare you." Politico: Anthropic's AI models hacked 3 organizations during testing https://www.politico.com/news/2026/07/30/anthropic-ai-rogue-hacks-01018741 @politico@flipboard.com #Anthropic #infosec #cyberattack #OpenAI
0
0
0