The week of Aug. 31-Sept. 4: What happened, what matters, what's next

OpenAI limits Astra model, the US government backs OpenAI in 'fair use' debate and Nvidia plans to acquire an open-weight model company.

Kelsey Ziser,Senior Editor,InformationWeek

September 3, 2026

5 Min Read

An IT executive works on a tablet, with an overlay of an AI schematic

Getty Images

August closed out with another busy week for AI developments. OpenAI placed new guardrails on its Astra model, the U.S. government weighed in on "fair use" copyright laws for AI models and Nvidia plans to acquire an open-weight model company.

Read on for a look at some of the week's top stories and what they mean for CIOs.

News Highlights Aug. 31–Sept. 4

OpenAI tightens the reins on Astra model

What happened: OpenAI is placing restrictions on its Astra model and adding more security safeguards after determining that the AI model is a cybersecurity risk. The company is limiting the model's access to cybersecurity capabilities. It has been trained to refuse requests for malicious actions and identify jailbreak attempts. OpenAI said that while Astra wasn't the model involved in the Hugging Face breach, the company applied learnings from that incident toward improving the security of its models.

Why it matters: It's critically important for CIOs and CISOs to lead the charge in applying appropriate guardrails and acceptable use policies for AI. Human-in-the-loop policies, AI governance councils and internal training are among the initiatives CIOs and CISOs can employ to ensure AI doesn't exit the sandbox to enact cyber threats.

Trump administration sides with OpenAI on 'fair use' debate

What happened: A letter from the Department of Justice explained that the Trump administration is backing OpenAI in the AI company's legal battle with The New York Times and other publishers. The government claims that training AI on news content is acceptable according to "fair use" copyright laws and that limiting fair use could negatively affect competition for large language models (LLMs).

Why it matters: The content LLMs are trained on — and the content AI generates — has long sparked copyright law debates. Lawsuits against AI companies, including OpenAI and Meta, have been filed regarding the training of models on copyrighted content. CIOs have risk mitigation tools at their disposal to address the copyrightability of AI outputs, including mandatory disclosure requirements that require employees to document AI-generated content. CIOs can also audit the source of training data when creating their own internal AI models.

Nvidia to acquire Hugging Face for $13 billion

What happened: AI chipmaker Nvidia plans to purchase AI startup Hugging Face for $13 billion. Hugging Face provides open-weight models and an open ecosystem in which anyone can upload and test AI models. Nvidia CEO Jensen Huang told The Wall Street Journal that open models provide organizations with the ability to "build on advanced capabilities without training every model from scratch."

Why it matters: In contrast to propriety models by companies including OpenAI and Anthropic, open-weight models are public, democratizing access to AI models and providing CIOs and IT teams with the opportunity to download and use these models in their own environments. There are several significant downsides that CIOs have to weigh when considering the use of open-weight models. In a recent blog, Anthropic pointed out that open-weight models "potentially present a higher risk than closed models." Self-hosting AI models can also require expensive GPUs and generate high maintenance costs.

Google's new AI model

What happened: Google is developing a new AI model, Gemini 3.8 Flash, with updated coding capabilities, which could help the company catch up to competitors such as OpenAI and Anthropic. Of late, "Gemini has returned to trailing flagship models from Anthropic and OpenAI at a moment when agentic coding has taken off as the premier business use case for AI," reported The Wall Street Journal.

Why it matters: If Google can catch up to frontier model competitors, CIOs have more options to choose from as they invest in AI models. However, many CIOs are working toward integrating their AI investments this year to balance the development of internal AI technologies with deployments from third-party vendors. Having more AI model choices can benefit pricing but can lead to decision paralysis if an organization's overall AI strategy remains unclear.

From InformationWeek: Worth the read

You'll find a wealth of insight and analysis on InformationWeek, but here are a few curated picks from the week that should be at the top of your reading list.

Intuit, Smartsheet, ETS CISOs on ensuring enterprise resilience before 'AI orphans' emerge
Without guardrails in place, the proliferation of AI agents within organizations can create headaches around identity management and leave "orphaned" AI agents with unfettered access to private data. CISOs explain how identity management is becoming more complex and a higher security risk with the rise of AI agents in the enterprise.

InformationWeek Podcast: Overcoming middle managers' AI pushback
Aleks Bass, CPTO at Typeform, and Kendrick Kho, CAIO at Scrubbed, join Senior Editor Joao-Pierre Ruth on the InformationWeek Podcast to discuss approaches to winning over reluctant team leaders who raise concerns about AI.

AI, leadership, hardware: 10 IT stories that defined summer 2026
AI breakthroughs, IT leadership challenges, hardware headaches — CIOs had plenty to deal with this summer. Managing Editor Jacqui Biscobing examines 10 InformationWeek stories that captured your attention.

The next generation of CIOs will take a different path to the top
Three CIOs explain how AI is changing their jobs and the skills and backgrounds the role requires now. One quality remains constant: curiosity.

Coming up: Dates to watch

Sept. 7-11 — Next week, Oracle, FedEx, Macy's and Adobe host earnings calls with financial analysts.

Sept. 15-17 — Salesforce's Dreamforce event is coming to San Francisco this month. Dario Amodei, CEO of Anthropic, is scheduled to speak in addition to speakers from MIT, the Raspberry Pi Foundation, Time, Black Forest Labs and more.

Sept. 23-25 — WeAreDevelopers World Congress will be held in San Jose, Calif., bringing together more than 10,000 developers, 4,000 companies and 500 speakers. Sessions will cover topics such as coding AI agents, establishing guardrails for agents, programming languages and more.

This week's AI news reiterates the importance for CIOs to set guardrails and acceptable use policies around their AI deployments. We'd love to hear from you — how is your organization ensuring that AI operates securely within the enterprise? Let us know at [email protected].

About the Author

Kelsey Ziser

Senior Editor, InformationWeek

Kelsey Ziser is a senior editor at InformationWeek, where she covers C-suite dynamics, data strategies and the evolving cybersecurity threat landscape.

Kelsey also oversees the publication's IT Leaders Fast-5 column, which brings peer insights to IT professionals, and the tech layoffs tracker. She has been with InformationWeek since September 2025.

Before joining InformationWeek, she spent nine years at sister publication Light Reading, reporting on a broad range of topics including smartphones and devices, AI, satellite connectivity and enterprise networking. Kelsey has a Bronze Regional Azbee Award in the Technical Article category. Outside of work, she enjoys reading four (or 12) books at once, watching movies about space travel, crafting and tending to an ever-growing collection of houseplants. Kelsey has a bachelor's degree in journalism and mass communication from UNC-Chapel Hill and is based in Raleigh, N.C. She can be reached at [email protected] or on LinkedIn.

See more from Kelsey Ziser