Oracle Cuts About 21,000 Jobs Amid AI Push
According to Reuters, Oracle's workforce has shrunk by about 21,000 employees (roughly 13%). The rollout of AI is cited as a factor behind the job cuts. As companies use AI tools to boost operational efficiency, examples of traditional roles shrinking are now surfacing even at major tech firms.
Oracle has long been a pillar of the cloud infrastructure and database business, and it appears to be accelerating AI-related investment while cutting existing operations staff. That said, not all of the reductions are necessarily driven by AI — some see them as part of a broader business reorganization. Either way, the impact of AI adoption on employment structures is reaching a scale that can no longer be ignored.
California Admits Using Unreported 'High-Risk' AI Systems
A CalMatters investigation revealed that California state agencies have been operating AI systems classified as 'high-risk' in risk assessments without filing the required reports. The state had previously pledged transparency around AI use in government, but in practice several systems slipped through the reporting requirements.
The issue once again calls into question the effectiveness of AI governance in government agencies. It highlights that setting rules alone is not enough — operational oversight and compliance with reporting obligations are essential. Similar challenges may be lurking in other states and countries.
The 'AI Identity Crisis' Facing Software Engineers
In a Business Insider article, a VC partner commented on the 'identity crisis' spreading among software engineers. With the rise of AI coding tools, engineers' sense of self — 'who am I, and what value do I provide?' — is being shaken.
As the act of writing code itself is increasingly taken over by AI, engineers are expected to bring higher-level skills such as design judgment, requirements definition, and quality assurance. The observation that many engineers caught in this transition are losing their sense of direction points to a structural shift across the industry.
The Blind Spot in AI Evals — Why the 'Linear Sales Email Failure' Gets Missed
An analysis by Tenure AI argues that many current AI evaluation (eval) methods fail to capture failure patterns that arise in real production environments. The specific example cited: an AI-generated sales email that contained content inappropriate for the context of Linear.
Many AI evals are run on generic tasks, but in actual business settings, failures are often context-dependent. This 'gap between evals and reality' becomes a critical risk factor when deploying AI systems to production. The need for evaluation methods that better reflect real-world context is coming to the fore once again.
Google Invests in Film Studio A24, Building an AI Research Partnership
According to the Wall Street Journal, Google is moving forward with an investment in A24, the independent film studio behind 'Everything Everywhere All at Once,' along with a partnership in AI research. It's drawing attention as a new form of AI adoption in the entertainment industry.
Google has been expanding into creative fields centered on its Gemini models, and a direct partnership with Hollywood — the front line of content production — suggests AI is moving into serious use in film and video production. That said, there are still many issues to resolve, including creators' rights and copyright concerns.
Gray Swan's Red Teaming — The Front Line of AI Security
In a Latent Space interview, OpenAI board member Zico Kolter and Gray Swan CEO Matt Fredrikson discussed the current state and challenges of red teaming in AI security. Gray Swan is a specialized firm that hunts for vulnerabilities in AI models, and they touched on a breakthrough known as 'Mythos.'
They emphasized that AI security is not merely 'cybersecurity powered by AI' but a distinct discipline in its own right, one focused on assuring the safety of the models themselves. In particular, evaluating model robustness against adversarial attacks is becoming an essential component of deploying AI in the real world.
---