MITRE ATLAS™ (Adversarial Threat Landscape for Artificial-Intelligence Systems) is a globally accessible, living knowledge base of adversary tactics and techniques based on real-world attack observations and realistic demonstrations from AI red teams and security groups. There are a growing number of vulnerabilities in AI-enabled systems, as the incorporation of AI increases the attack surface of existing systems beyond those of traditional cyber-attacks.

MITRE | ATLAS (Adversarial Threat Landscape for Artificial-Intelligence Systems) is a globally accessible, living knowledge base of adversary TTPs based on real-world attack observations and realistic demonstrations from AI red teams and security groups.

Increased context windows in LLMs are great. But here is a jailbreaking technique that exploits exactly that. It uses many-shot learning: hundreds of demonstrations to steer model behavior. Many-shot Jailbreaking (MSJ) is a direct long-context attack that exploits multiple instances of undesirable behaviors. The effectiveness of these attacks increases according to a power law to the number of examples provided! Despite altering the format or content of the prompts, mitigating these attacks remains challenging, suggesting that very long contexts unveil a new vector for adversarial strategies against LLMs. The study uncovers that models of larger sizes tend to be more susceptible to many-shot jailbreaking, complicating the task of safeguarding them without diminishing their capabilities. Blog post: https://lnkd.in/ex4TvDsD Paper: https://lnkd.in/ef2DWa7H

#ICYMI, our top OSINT videos from March showcased weekly demonstrations on how to utilize the Forensic OSINT Chrome Extension along with another OSINT tool. These are the OSINT tools we highlighted: ✅ Wayback Machine 🕰️ ✅ Add-on: Fake News Debunker 🕵️♂️ ✅ Whoxy 🔍 ✅ Whatsmymame 🤔

Psst, did you know that here at Ace Computers we are a leading technology provider for the Forensics, Intelligence, and Data Extraction/Analysis communities? Not only do we excel in HPC and other computing solutions, but our technology solutions go above and beyond for our Law Enforcement and Federal Agencies. #AI #Intelligence #HPC #Forensics

Data Scientist Michael Ryaboy discusses KDB.AI's Temporal Similarity Search (TSS) and how it accelerates real-time data analysis in this recent live event. Michael shows how TSS helps developers conduct near real-time similarity searches on rapidly evolving time-series data. Watch now for code demonstrations of Transformed TSS and Non-Transformed TSS. Stream the entire event: https://lnkd.in/enHgwHVX  #KX #KDBAI #TSS #DataAnalysis #DataInsights #RealTimeAnalytics

🔗 Exciting news! Check out this insightful blog post on "Any-point Trajectory Modeling for Policy Learning," which presents a novel framework called Any-point Trajectory Modeling (ATM). This framework leverages video demonstrations to pre-train a trajectory model, enabling the prediction of future trajectories of arbitrary points within a video frame. With minimal action-labeled data, this approach facilitates the learning of robust visuomotor policies. Dive into the details and explore the visualizations and code at: https://bit.ly/3H1Ty6r.

Seven minutes are all it takes for the fastest adversary to break out and start moving laterally. Experience a live demonstration of CrowdStrike Falcon LogScale to learn how to triage, investigate and respond to STARDUST CHOLLIMA. https://lnkd.in/ei4kYqxb