September 9 news:
- Anthropic endorses SB 53. Anthropic signalled its support for the new ‘trust-but-verify’ California bill on governing AI systems, given the difficulty of addressing AI safety at a federal level in the United States. Among other things, the bill requires companies like Anthropic to develop and publish safety frameworks, which describe how they manage, assess, and mitigate catastrophic risks—risks that could foreseeably and materially contribute to a mass casualty incident or substantial monetary damages and release public transparency reports summarizing their catastrophic risk assessments and the steps taken to fulfill their respective frameworks before deploying powerful new models. Sounds like bare minimum requirements. Anthropic
- DataKrypto joined NVIDIA Inception. DataKrypto announced it has joined NVIDIA Inception, an NVIDIA program that makes resources and expertise available to startups. For DataKrypto, the intent is to accelerate the delivery of Confidential AI: encryption-based data, model, and workflow protection solutions that give customers stronger defenses against today’s evolving threats. It’s FHEnom for AI is the focus. DataKrypto
- EMA, CISA and Zero Networks on microsegmentation. EMA’s latest research on zero trust and microsegmentation found that 96% of respondents believe microsegmentation to be extremely or very important for cyber defense, with key benefits including instantly quarantining threats, halting lateral movement, and meeting compliance requirements. In conjunction, CISA has recently advocated for the wider adoption of microsegmentation, a journey that Zero Networks says its ready to assist with. BusinessWire
- BeyondTrust on solving the hiring fraud problem. With the rise in hiring fraud, particularly the types initiated from North Korea, the question is how to deal with the threat. BeyondTrust’s proposed solution is zero standing privileges, encompassing no always-on access by default, just-in-time and just-enough-privilege, and auditing and accountability. This approach closes the gap left by the castle problem. It ensures attackers can’t rely on persistent access, while employees can still move quickly through their work. Done right, a ZSP approach aligns productivity and protection instead of forcing a choice between them. See BeyondTrust Entitle. Hacker News