BIG TECH

New events in OpenAI-Apple lawsuit raise the risks

In the span of two days, OpenAI and Apple each made their next moves, and the rift between them grew wider.

Apple kicked off the week by filing a motion for a preliminary injunction against OpenAI on Monday, asking the court to bar OpenAI and two former Apple employees from accessing, acquiring, using, or disclosing Apple's alleged confidential information on the grounds of "trade secret theft." Apple also filed a concurrent motion for expedited discovery, seeking a faster production of evidence from OpenAI, arguing that time is of the essence.

"Further, Defendants’ possession and use of Apple’s proprietary and secret information is ongoing," said the company in the motion. "And, if Apple’s trade secrets are used and spread throughout OpenAI, or get embedded in its products and operations, the damage cannot be undone.”

OpenAI clapped back with a blog post on Tuesday, posting a series of text messages and emails that the company points to as evidence absolving it of Apple's accusations. The AI lab also claims that it tried to work with Apple on the issue before the lawsuit was filed, but instead Apple is "trying to change their narrative, including by making vague accusations about other former employees." 

The choice to make the receipts public is not a legal play; rather, by bringing their side of the story to the court of public opinion, OpenAI is likely hoping that external pressure and scrutiny will make Apple retract its position. The company asserts that it does not hold or want Apple's proprietary product knowledge. 

"Apple’s request for a preliminary injunction is both based on false information and completely unnecessary because we do not have, nor want, any of their trade secrets," said OpenAI. 

This is all happening against the backdrop of the original lawsuit, presented by Apple on July 10, which included an abundant amount of evidence supporting claims that former Apple employees, including Tang Tan, who worked at Apple for 24 years and is now OpenAI's chief hardware engineer, and Chang Liu, who worked at Apple for eight years, stole insider knowledge to work on Jony Ive's new device being developed with OpenAI.

While it is Apple that is claiming irreparable damage, OpenAI has a lot to lose. For one, time is of the essence in the AI space, and until this is resolved, which could take years, the company could be stymied in its plans to release its first AI hardware devices. As for the first device, it is rumored to closely resemble an Amazon Echo smart home speaker, something different from Apple's rumored upcoming AI smart glasses and smart pin, which may work in OpenAI's favor in proving originality. Regardless of the outcome, this is shaping up to be a long and complex process. Another factor is the company's upcoming IPO, the success of which depends heavily on investor confidence, and uncertainty or a negative outcome from the Apple lawsuit could impact that.

IN PARTNERSHIP WITH LAMBDA

Cut your AI training costs by 25% or more

Most large-scale AI training runs use less than half the computing power they're paying for. Lambda's team found the root causes and built a reproducible framework that boosted efficiency by over 25%, without changing the model itself.

Lambda’s whitepaper shows you how to address:

• Memory inefficiencies silently inflating your costs

• Training configurations that aren't making full use of your hardware

• Bottlenecks that slow down GPU communication

Get the whitepaper.

GOVERNANCE

Mistral bets on AI's missing safeguards

Though successful AI deployment requires guardrails to keep models from going rogue, actually implementing those guardrails is often arduous. Mistral is working to change that.

On Tuesday, the French AI lab launched Shieldstral, a 3 billion multimodal safety classifier, which can also be taught to act as a content moderation or guardrail model that can control what the AI system is able to do. Unlike traditional guardrail models, Shieldstral accepts natural language to write policies at inference time, circumventing the need for retraining and making this process more seamless. 

When users input a natural language query describing a safety concern, Shieldstral produces a single continuous safety score. This gives users a more detailed metric they can use to make informed decisions about the level of risk and set their own thresholds for what to allow or block, rather than locking deployers into a fixed binary safety check.

Mistral claims it outperforms models up to seven times its size, beating OpenAI's GPT-OSS-Safeguard, a 20-billion-parameter model, on benchmarks results for text safety. The advantage of a smaller model is being able to run it using less compute, with Shieldstral able to run on as little as a 16GB GPU, which also makes it far more cost effective. 

The model is open-weight and available under Apache 2.0. The move is on-brand for Mistral, which has typically followed an open-source approach and recently joined the Open Secure AI Alliance for AI Safety and Security, a coalition launched this month to build and share open tools.

There's no denying that Mistral isn't currently in the same tier as frontier labs like Anthropic or OpenAI, whether measured by model capability or market visibility. But the company has quietly excelled at something else: Rather than pouring every resource into the race for the next frontier model, it has focused on building practical solutions to the unglamorous problems scattered across the AI stack. Shieldstral is a clear example of this. It doesn't chase headlines the way a new flagship model does, but tools like it are exactly what move the field forward in practice. If more companies adopted this approach, the entire industry would be better for it, saving money and deploying safely rather than chasing state-of-the-art capabilities and multi-trillion parameter, power-hungry models.  

IN PARTNERSHIP WITH ROKT

What do your customers already know before they arrive?

What has this customer already figured out before arriving here?

That's the new question businesses need to ask as AI reshapes how customers research, compare, and decide. Many are arriving with their minds already made up, and the businesses that recognize this will have an edge.

Rokt's latest guide breaks down what this shift means and how to respond to it.

Read the full guide.

WORKPLACE

How AI is rewriting entry-level work

AI may be causing the job market to splinter. 

As companies adopt AI, UK employers are increasingly prioritizing experienced workers with AI skills, raising the bar for early-career professionals, according to new data from Indeed Hiring Lab. The research finds that graduate job postings in the UK have fallen to their lowest share since 2020, even as AI-exposed occupations like software development, IT, and engineering outperform much of the broader labor market.

"The recent rise in software development hiring has focused on senior and AI-specialist roles, suggesting that employers are prioritizing experience and AI fluency over volume hiring of junior talent," Jack Kennedy, senior economist at Indeed Hiring Lab, told The Deep View. "That makes the skills mix more important than ever."

Overall hiring remains subdued, with U.K. job postings 11% below where they stood at the beginning of 2026. But the slowdown is reversed when AI skills enter the equation. 

While hiring remains weak across many white-collar occupations, job postings that explicitly reference AI continue to surge, according to Indeed. Job postings for HR, management, marketing, and finance that mention AI are growing even as overall postings in those fields decline.

"We're seeing the relationship between AI exposure and hiring demand has flipped, with occupations most exposed to AI now seeing stronger hiring growth on average," Kennedy said. "This is true not just for software development, but also other AI-exposed occupations."

The data suggests employers are looking to hire for experience and AI fluency. As AI increasingly handles routine coding tasks, Kennedy said companies are placing greater emphasis on prompt engineering, code review, and quality assurance alongside traditional technical expertise.

"For jobseekers, particularly those in the early stages of their career, investing in that skills combination will be the most effective way to adapt to a market that is being rapidly reshaped," Kennedy said. "Demonstrating genuine AI competency alongside strong technical, creative and problem-solving skills is increasingly what separates candidates who progress from those who don't."

Indeed's findings complicate the narrative that AI is coming for jobs. Since the introduction of generative AI, economists and business leaders have speculated that AI will automate the tasks of white-collar workers, rendering many roles obsolete. Software engineers, in particular, have gotten the most heat. While studies do suggest coding roles are highly exposed to automation, that doesn't necessarily mean they're at risk of disappearing. Instead, AI is changing the work itself. For developers, that means spending less time writing code and more time reviewing, polishing, and validating AI-generated outputs. The same dynamic is playing out in fields like marketing and advertising, where professionals increasingly use their expertise to refine AI-generated work rather than create it from scratch. The shift suggests AI literacy is becoming increasingly important for employment. That’s especially true for college graduates entering a labor market where entry-level opportunities appear to be shrinking. Still, it raises questions about the quality of the work itself. Editing AI-generated outputs may not be as satisfying as creating something original, and whether workers will embrace that tradeoff remains an open question.

LINKS

• Anthropic signs $10 billion, six-year compute deal with Volta Infra

• White House claims AI evaluation framework done but withholds details

• HappyRobot raises $150 million Series C at a $ 1.2 billion valuation

• World Bank says AI poses less job threat in developing economies

• Medical AI tools benefit clinicians more than non-experts, MIT study finds

• Linux launches Tokenomics Foundation to standardize AI economics

• Gemini Notebook: The more powerful Notebook experience rolled out to all Pro users

• MiniMax-H3: Now the #1 open model in Video Arena

• Gemini Spark: can now tap into Google Chrome’s auto-browse feature

• Cursor: agents are now 20-30% more token efficient, according to the company

• Pixi Inc: AI analyst 

• AppZen: AI Engineer

• Campfire: AI Engineer

• Symmetry AI: Applied AI Engineer

GAMES

POLL RESULTS

Do you think that frontier AI labs need to slow down the pace of development?

Yes (42%)
Somewhat (21%)
No (29%)
Other (8%)