AI Agents in Cyberattacks, Inception's Mercury 3, and Anthropic's Singularity Outlook
Security startup NorForge exposes AI hacking tool dashboards, while the US Department of Homeland Security bans DeepSeek in federal agencies. Inception launches the Mercury 3 diffusion model, and Anthropic's Jack Clark discusses the onset of the AI singularity.
Podcast В· 2 min
AI Agents Used in Cyberattacks; DHS Bans DeepSeek
Security startup NorForge has identified exposed control panels for an AI hacking tool called Cairn, which appears to be used by criminals to automate cyberattacks. The report highlights that these agents have targeted over 1,100 projects, with a 10% success rate, costing roughly $50 per successful break-in. Victims identified include John Deere, SharkNinja, and Cummins. Separately, the US Department of Homeland Security has formally ordered federal civilian agencies to remove DeepSeek products from their systems, citing security concerns, and extended the ban to third-party tools built on the Chinese lab's models.
Inception Launches Mercury 3; Anthropic Nears New Compute Deal
AI startup Inception has released Mercury 3, a diffusion-based language model that the company claims offers high-speed generation, reportedly reaching 1,068 tokens per second on NVIDIA GPUs. Meanwhile, Anthropic is reportedly nearing a second $45 billion compute deal with Nscale, following their initial agreement just over a month ago. Additionally, Tenstorrent has raised $1.5 billion at a $6.2 billion valuation to scale its open-architecture AI chips, positioning itself as a competitor to Nvidia.
FLUX 3 Max and MiniMax M3.1 Flash Released
Black Forest Labs has released FLUX 3 Max, a new text-to-video model that reportedly generates 10-second clips in approximately 12 seconds, offering significant speed improvements over previous iterations. In the model space, Chinese AI lab MiniMax has launched M3.1 Flash, a new model that has shown strong performance in coding and computer-use tasks while being positioned as a low-cost, high-efficiency option for running AI agents.
Anthropic Co-founder on AI Singularity
Jack Clark, co-founder of Anthropic, stated at the IMF and World Bank annual meetings that the AI singularity—the point at which AI-driven progress outpaces human ability to follow—has officially begun. Clark cited AI's rapid improvement on expert-level benchmarks, such as Humanity's Last Exam and FrontierMath, as evidence. He also noted that Anthropic CEO Dario Amodei has discussed potential limits on AI's intelligence ceiling, suggesting that there is only a small window of time to address these developments.