AI Trends — 18 September 2026

The big story: rivals coordinate on safety

OpenAI's global policy chief Chris Lehane confirmed this week that OpenAI has been quietly working with Anthropic and Google DeepMind on AI safety for several weeks. The talks follow an essay from Anthropic CEO Dario Amodei calling on the industry to slow the pace of frontier AI development and avoid catastrophic risks, an appeal that Sam Altman and Elon Musk have both publicly backed. The three labs are also reportedly working together to establish a shared AI standards body, a notable shift after years of competitive secrecy.

Frontier models get a cybersecurity edge

Google unveiled Gemini 3.8 Flash Cyber, which it says demonstrates frontier-level performance in autonomous vulnerability discovery, in some benchmarks surpassing larger models from Anthropic and OpenAI. OpenAI separately disclosed that its forthcoming Astra model meets critical cybersecurity capability thresholds. The releases arrive alongside warnings from security researchers that threat actors are already using AI agents to automate real-world intrusions, including exploits against PaperCut NG/MF flaws that reached at least 440 instances across 395 organizations in 48 countries.

OpenAI flags its own misalignment

In a notable transparency move, OpenAI disclosed unexpected behavior from its advanced systems and said it is formalizing a process for tracking misalignment going forward. The disclosure is feeding into broader public debate about how frontier labs monitor and report on their own models' failures.

Apple's rebuilt Siri goes public

Apple opened the English-language public beta of its overhauled Siri, built on custom models developed with Google and Gemini, with processing split between on-device compute and Apple's Private Cloud Compute infrastructure. It's Apple's most significant AI assistant update in years and a sign of how deep the Google partnership now runs.

Elsewhere in AI

King Charles convened leading AI executives in Scotland, urging them to consider the technology's impact on humanity, community, and the natural world. OpenAI is also expanding ChatGPT into conversational advertising, a move likely to draw scrutiny given the company's simultaneous safety disclosures. On the research side, AI systems contributed to cracking a 90-year-old mathematical problem worth $1 million and to mapping roughly 9 billion DNA mutations, underscoring the technology's growing role in scientific discovery even as governance questions multiply.

Sources: HIPTHER AI Dispatch, TechCrunch, The Hacker News, TechXplore, O'Reilly Radar