Anthropic: Artificial intelligence company Anthropic has released a new cybersecurity model known as Mythos. In an effort to stop the tool from being misused, Anthropic has only made Mythos available to around 40 companies, including Apple, Google, Microsoft, Nvidia, Amazon Webservices, CrowdStrike and Cisco. Through the initiative, known as Project Glasswing, company partners will receive $100 million in usage credits to employ Mythos as part of their cybersecurity defense.  Anthropic has said it will also donate $4 million to open source security organizations to allow the makers of these widely used platforms to better respond to new threats in the age of AI. Anthropic pledges to release information regarding its findings through this research initiative, as well as recommendations for evolving security practices in the AI era. Already, Mythos has been used to identify a 27-year old vulnerability in OpenBSD (a security focused operating system) and a 16-year old flaw in FFmpeg (commonly used for audio and video encoding/decoding). Anthropic contends that Mythos “reveals a stark fact: AI models have reached a level of coding capability where they can surpass all but the most skilled humans at finding and exploiting software vulnerabilities.” 

Meta: On Wednesday, Meta (formerly Facebook) introduced Muse Spark, a new artificial intelligence model. Muse Spark is the first model from Meta Superintelligence Labs, which was created last year. Notably, this model will introduce “Contemplating Mode,” which utilizes multiple agents running in parallel to result in a performance boost. The model scored a 50.2% on Humanity’s Last Exam, which is a test designed to assess the capabilities of artificial intelligence. With 2,500 graduate-level questions across a variety of subjects, Humanity’s Last Exam typically sees low scores (44.4% for Google’s Gemini 3.1 Pro and 42.7% for OpenAI’s GPT-5.4 Pro). During safety testing, a third party company found that Muse Spark exhibited the highest rate of evaluation awareness of any model that it had tested. The model was regularly able to recognize that it was being tested regarding its responses to unsafe situations, but there was no evidence to suggest that this impacted behavior. 

LAPD Hack: Details have become clearer regarding the Los Angeles Police Department hack that occurred last month. 337,000 files totaling 7.7 terabytes have been leaked, all of which came from a digital storage system operated by Los Angeles City Attorney’s Office. Internal Affairs reports, which are rarely published, were also included in the breach. In a statement, a spokesperson for the city attorney’s office said that “the information was self contained” and that the “investigation is continuing to determine what information was present.”