OpenAI Announces Cybersecurity-Focused "Daybreak" — Unveils GPT-5.5-Cyber and Codex Security

On June 22, OpenAI announced "Daybreak," a new suite of tools aimed at improving organizational security. The centerpiece products are "Codex Security," which discovers, verifies, and fixes vulnerabilities at scale, and "GPT-5.5-Cyber," a cybersecurity-specialized model.

At the same time, the company also launched the "Patch the Planet" initiative to support open source maintainers, offering a system that combines AI with expert review to accelerate vulnerability response in OSS.

Applying AI to security is a hotly contested field, and OpenAI's full-fledged entry could spur competition across the entire industry. That said, AI models themselves are a double-edged sword that can also be repurposed for attacks — something that warrants continued vigilance.

Google DeepMind Invests $75M in Film Studio A24 — AI Filmmaking in Hollywood

Google DeepMind has partnered with A24, the studio known for indie films, investing $75 million toward building AI filmmaking tools. The use of generative AI in Hollywood remains a contentious area, with ongoing debates over copyright and jobs, but a major tech company investing directly in a studio is a new pattern.

This move signals that practical, real-world adoption of AI tools in film production is taking shape. At the same time, many challenges remain to be solved, from protecting creators' rights to ensuring content quality.

Google Overhauls the Default Gemini API with the "Interactions API"

Google DeepMind has switched the default interface for Gemini models and agents from the conventional generateContent API to the "Interactions API." The new API adopts a simpler schema that uses typed steps instead of a role-based structure. Going forward, agent features will be provided exclusively through this new API.

While this means migration costs for developers, it could also bring clarity to the paradigm for building agents. It's evident that Google is in a hurry to standardize the agent layer.

Anthropic and Micron Partner on AI Memory Co-Design — Micron Also Invests in Series H

Anthropic has entered a multi-year partnership with memory maker Micron. Micron is investing in Anthropic's Series H round and will supply memory for the company's Claude infrastructure. Anthropic co-founder Tom Brown positions memory as "essential for training and inference."

As memory grows increasingly critical in AI infrastructure, vertical integration between model developers and hardware makers is accelerating. Some critics argue, however, that these circular investment relationships are fueling a bubble.

Amazon Begins Testing Alexa+ in India — with Hindi Support

Amazon is expanding the rollout of its AI assistant "Alexa+" to India, inviting users to test a Hindi version. India is a market with a massive population and a multilingual landscape, and the success of localization will hold the key to Alexa+'s global expansion.

A Warning on AI Training Data Quality — "In the End, They're Relying on Chatbots"

According to reporting by New Scientist, it has come to light that people involved in training AI models are outsourcing the data labeling and quality control work itself to chatbots. Structuring AI output as training data for other AI carries the risk of quality degradation and "model collapse."

Some point out that the rapid spread of AI is eroding incentives for humans to create high-quality data, once again calling into question how training data is sourced.

---