Prompt injection against spam classifiers: TF-IDF models scored 0% because a bag of words has no idea what an instruction is. Llama 3 missed every injected spam message.

10 msgs/condition, so read the intervals. Preprint:

doi.org/10.5281/zeno...

#AdversarialML #LLMSecurity #PromptInjection

#LLM #LLMs #LargeLanguageModels #ArtificialIntelligence #AI #ContentModeration #DigitalSafety #TrustAndSafety #AIModeration #NLP #NaturalLanguageProcessing #ModelEvaluation #Benchmarking #AdversarialML #SocialMedia #OnlineHarms #ResponsibleAI

norecognition.org

#AI #Privacy #CyberSecurity #ComputerVision #AdversarialML

Loved being back in Berlin for @wearedevelopers.bsky.social World Congress. Great to catch up with friends and chat with so many amazing people!

🎉 Recording of my talk “Confuse, Obfuscate, Disrupt” is out now: bit.ly/3I5OEJU

#AdversarialML

Spent a week poisoning my own RAG pipeline through the document corpus. not the prompt. the documents themselves.

32 vectors. 19 successes.

your "safe" RAG pipeline might just be confidently wrong.

corrupted.io/2026/04/24/P...

#AIsecurity #RAG #LLM #infosec #adversarialML

New article!
📖 Discover SecML-Torch, an open-source Python library from UNICA’s sAIfer Lab, designed to advance research in Adversarial Machine Learning (AML) and evaluate ML model robustness.

🔗 Read the full article: coevolution-project.eu/secml-torch-...

#AI #AdversarialML #Cybersecurity

🥊When the Algorithm Fights Back – Continuous adversarial ML testing is the frontline against AI-targeted cyberattacks, hardening models before threats exploit them. #CyberSecurity #MachineLearning #AIThreatDefense #AdversarialML #TheCyberLens

thecyberlens.com/p/when-the-a...

📢 We are excited to share that 4 new papers acknowledging #CoEvolution have been accepted for publication! 🎉

📝 Topics include adversarial robustness, federated learning & 3D perception.

👉 Explore them on our website: coevolution-project.eu/publications

#AI #AdversarialML #Cybersecurity #Research

Adversarial Clothing and GaP Patches Targeting Clearview AI and Amazon...

https://www.schneier.com/blog/archives/2026/08/adversarial-clothing-designed-to-fool-facial-recognition-systems.html
https://flagthis.com/tldr/5858

##AmazonRekognition ##ClearviewAI ##AdversarialML ##ComputerVision ##Privacy

#ControlTheory for #adversarialML at #AAAI2021 paper similar to our #CVPR work w/ @ArashRahnamaPhD @AndreNguyen16 is On Lipschitz Regularization of Convolutional Layers using Toeplitz Matrix Theory

This was another great ep following the one with Simon Willison about finding the boundaries of LLMs oxide-and-friends.transistor.fm/episodes/…

#podcast #machineLearning #LLM #adversarialML

Researchers demonstrate a timing‑based adversarial attack that leaves AI output unchanged but adds delays or contradictory cues, increasing error rates in decision tasks. Read more: https://getnews.me/human-factors-redefine-adversarial-analysis-for-ai-decision-systems/ #adversarialml #humanfactors

A Bi‑Task Adversarial Attack can lower detection confidence and shift monocular depth; printed‑patch tests show it works in the real world. Read more: https://getnews.me/bi-task-adversarial-attack-threatens-object-detection-and-depth-estimation/ #adversarialml #objectdetection #depthestimation

The preprint, submitted September 2025, shows a batch estimator with exponential error decay as Markov‑parameter order k rises, even with ~1/k attack probability. Read more: https://getnews.me/batch-and-streaming-estimators-boost-system-identification-against-attacks/ #sysid #adversarialml

Adversarial robustness is key for reliable AI!🛡️Exploring new defense strategies. [URL] Awesome Drawing tools for Neural Net Architecture https://john-smiths-is-me.github.io/ #adversarialML #security #AI #ICML #NIPS

Adversarial robustness remains a critical challenge! 🛡️ Exploring new defense strategies & understanding attack surfaces is key to trustworthy AI. 🤔 #AdversarialML #Robustness #AIsecurity #Defense
Awesome Drawing tools for Neural Net Architecture https://john-smiths-is-me.github.io/

#AdversarialML #Robustness #MachineLearning #ICLR2025

Remarkable admission: top security researcher Carlini finds Claude better at vulnerability discovery than himself. He just uncovered a 2003 Linux buffer overflow no one caught for 20 years. #security #infosec #adversarialml

https://bymachine.news/carlini-claude-security-researcher-linux