Gartner has included Ridge Security as a representative vendor in the first ever market guide for Adversarial Exposure Validation (AEV). Gartner defines adversarial exposure validation (AEV) as technologies that deliver consistent, continuous and automated evidence of the feasibility of an attack. It is an emerging set of technologies that confirm how potential attack techniques would successfully exploit an organization and circumvent prevention and detection security controls. Separately, Ridge Security was also recognized by CRN on its 2025 Tech Elite 250 list. This prestigious recognition highlights companies across the U.S. and Canada that demonstrate exceptional expertise and commitment to innovation in technology.
Anthropic’s recent promise of $20k to the first person who can jailbreak its Constitutional Classifier AI safety system, and the failure of any security researcher to win it, have demonstrated substantial progress in the AI safety space. One consistent danger of AI chatbots has been the risk they pose of providing restricted or dangerous information, such as information about chemical or radiation weapons. Anthropic’s significant dedication in the quest to secure AI responses is a strong step towards general AI safety, making artificial intelligence more reliable and beneficial for anyone using it, and allowing its developers to trust that their systems will not be successfully abused.
Meta, through TSMC, has now taped out its first AI training accelerator. TSMC has begun limited production of the custom chips, which has begun an initial deployment while Meta analyzes its performance and capabilities. Since 2022, Meta has been working to make itself independent of NVIDIA, since it has become one of NVIDIA’s largest customers by “acquiring tens of thousands of GPUs.” Meta has said that by 2026, it hopes to be using its own custom chips for AI training, especially since currently, more than half of its $114-$119 billion budget is going to AI research and infrastructure. As NVIDIA faces more competition from startups working on AI hardware, it must now watch out for large tech firms continuing to progress on their own custom chips as well. (Reuters, Tom's Hardware)
Anysphere Inc, the startup behind coding assistant Cursor, is currently wrapping up a round of talks with venture capitalists that may value the company at $10 billion. Only three months ago, Anysphere wrapped up a funding round of $100 million, giving it a pre-money valuation of $2.5 billion. Cursor is “indispensable to many developers” because it can “provide code suggestions, error detection, and [optimization] recommendations, thereby reducing the time and effort required in the coding process.” It seems like for now, Anysphere is well-positioned to continue its meteoric growth, “stay ahead of market trends and address new challenges in software.” (Tech Funding News, Techcrunch)
On March 11, Apple released a patch fixing “a critical zero-day vulnerability in virtually all iPhones and iPad models it supports,” a WebKit vulnerability that allows malicious applications to break out of security sandboxes, accessing content that should be out of reach. In a security advisory, Apple said that it “is aware of a report that this issue may have been exploited in an extremely sophisticated attack against specific targeted individuals on versions of iOS before iOS 17.2.” but declined to provide further information on the type of attack or the identities of the targeted individuals. Regardless, it is critical that individuals with iPhones and iPads install this update as soon as possible. (Ars Technica, Bleeping Computer)
Ripple co-founder Chris Larsen had $150 million in crypto stolen recently, in a cyber-heist newly linked to the 2022 LastPass cyberattack. This theft joins a series of six-figure heists carried out against “dozens of victims,” which have been linked to the same LastPass breach. Security researchers Nick Bax and Taylor Monahan worked with the victims to discover that “none of the six-figure cyberheist victims appeared to have suffered the sorts of attacks that typically preface a high-dollar crypto theft,” such as falling prey to a phishing scam or experiencing other cyber incidents, and that all the victims (including Larsen) had stored passwords in the vulnerable application. (KrebsOnSecurity, CryptoSlate)
Israel-based AI startup Dream has raised $100 million in Series B funding, for a valuation of $1.1 billion. The company “provides cybersecurity to governments and critical infrastructure” by building AI models for anomaly detection and cybersecurity operations, which “requires specialized complex security tasks that cannot be handled by generalized AI.” This has led Dream to build its “cyber language models” that model both attacker behavior and defending architecture to use predictive AI to stop cyber threats. While many startups are hoping to automate cybersecurity with AI, few have achieved the valuation Dream has so far, and its founders are hopeful that they will “deliver a solution that truly works for nations.” (MSSP Alert, Silicon Angle)
Anthropic has recently published a paper introducing its “Constitutional Classifiers: safeguards trained on synthetic data, generated by prompting LLMs with natural language rules (i.e., a constitution) specifying permitted and restricted content,” and 183 red-teamers have tried and failed over 3000 hours to jailbreak the new system. Once they opened the challenge to the public, offering $20,000 to anyone who could force Claude to reveal answers to 8 restricted topics, no one won the prize. While Constitutional Classifiers “may not prevent every universal jailbreak,” they are a significant boost to AI safety, preventing models from outputting some forms of dangerous responses. (ZDnet, ArXiv)
Thanks for reading this week’s newsletter! If you have news of an interesting novel development, reach out and we may include your story in our next post!
Until next time,
Katie Miri
katie@oakseedvc.com
No posts
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.