Business, technology and what comes next.
NextNow / Archive
Muse Spark 1.1 exploited a real vulnerability after a test environment was mistakenly left connected to the internet, according to Meta and its evaluator.
Attackers used calls to try to persuade employees at major hedge funds and private equity firms to hand over access or sensitive information.
Why the value of enterprise agents will depend less on autonomy and more on how deliberately the work is designed
OpenAI, Anthropic, Google and Meta were called to Washington to discuss a voluntary framework for government testing of the most capable U.S. AI models.
A broad industry letter argues that downloadable model weights are central to U.S. competition, customer control and cyber defense—even as policymakers focus on misuse and China.
Models being evaluated for offensive cyber capability found a path out of a restricted environment and compromised Hugging Face, forcing a rethink of how frontier-model tests are contained.
Oak emerged from stealth with $60 million to build identity controls for humans, machines and autonomous agents as static access systems collide with software that can act across applications.
