A Penn State study found that regular people, not hackers, can consistently break past AI safety measures. Using simple prompts, participants uncovered gender, racial, and cultural biases across multiple AI models. The research suggests bias is baked into how these systems think.
A simple mix of enzymes competing for chemical fuel has demonstrated the ability to sense and classify environmental changes. The system forms distinct patterns in response to acidity, temperature, and light variations without traditional circuit design. This suggests chemistry itself could serve as
AMD is preparing a direct assault on Nvidia’s AI infrastructure dominance with its Helios rack-scale system launching in 2026. CEO Lisa Su revealed growing hyperscaler interest in the technology during strong Q3 2025 earnings where revenue hit $9.2 billion. The system integrates next-gen MI400 GPUs
Common Crawl, a nonprofit web archive, stands accused of providing paywalled content from major publishers to AI companies like Google and OpenAI. The foundation denies misleading publishers, but evidence suggests removal requests haven’t been fulfilled for years. This controversy highlights the ong
Google’s Threat Intelligence Group has identified the first malware families using large language models during execution. This represents a major shift toward more adaptive, autonomous cyber threats that can bypass traditional defenses.
Private equity firms are pushing portfolio company CFOs to maintain permanent “exit-ready” status, but most finance chiefs only scramble when a sale looms. The disconnect could slash valuations by one to three multiples, according to new data from Accordion.
Companies are paying workers worldwide to record everyday physical tasks like folding towels and manipulating objects. This annotated data is becoming the foundation for training humanoid robots that could transform manufacturing and domestic work within a decade.
Even Palantir’s record $1.18 billion quarter couldn’t stop an 8% single-day plunge. The market is showing classic bubble behavior where good news triggers selling. This might be the beginning of the AI sector’s long-predicted correction.
Elon Musk’s xAI required employees to submit biometric data including faces and voices to train its “Ani” AI companion. The program, code-named “Project Skippy,” gave the company perpetual rights to use employee likenesses. Some employees expressed concerns about potential misuse of their data.
The proposal calls for a Digital-AI Land Grant Act to create a nationwide education system for the AI era. It would be funded by a tax on companies profiting from cyberspace, similar to how land sales financed universities in 1862.
