Apple-OpenAI Lawsuit, Ransomware Attack: DesignRush AI Roundup

Our analysts track the weekly developments reshaping AI and how it reaches users. Brands building AI products can partner with vetted AI companies to bring their ideas to life.

Three stories moved the space this week.

Apple sued OpenAI, claiming the latter lifted iPhone hardware secrets through more than 400 former Apple employees.

Chinese open-source models climbed from about 5% of U.S. enterprise AI tokens in early 2025 to a weekly peak of 46%.

And an AI agent ran a full ransomware attack on its own, with no human operator at any step.

Here's a closer look at what happened and what each story means for the brands and agencies relying on these tools.

Apple Sues OpenAI Over Trade Secrets

Apple filed suit against OpenAI in federal court on July 10, alleging a coordinated scheme to lift iPhone hardware trade secrets through former employees.

More than 400 former Apple employees now work at OpenAI, and Apple says the practice runs "at every level" of its hardware operation.

The court filing names Tang Tan, OpenAI's chief hardware officer and a longtime Apple veteran, as a central figure.

He allegedly used internal Apple code names in interviews to pull more confidential information from candidates.

Bloomberg reported that the suit gives anyone weighing an OpenAI offer real litigation risk to consider.

OpenAI said in a statement that it has "no interest in other companies' trade secrets."

It is also still reportedly on track to ship its first hardware device in 2027.

With OpenAI weighing an IPO this year, a trade-secrets lawsuit is exactly the risk factor investors will read closely.

Chinese AI Models Are Winning on Price

Every SaaS budget eventually hits the same fork, and AI just reached this point.

The choice is to either stick with the market leader or switch to something cheaper that does nearly the same job, and the second option is winning.

Chinese models now run up to 46% of the tokens on OpenRouter, which lets developers reach many AI providers in one place.

This figure is up from an 11% average the year before, more than a fourfold jump.

AI startup Lindy also moved all its traffic from Claude to DeepSeek in June.

Founder Flo Crivello said the switch cut costs sharply while improving performance on core tasks.

Open-source Chinese models can be "60% to 90% cheaper" than leading U.S. labs, an analyst told CNBC.

In raw capability, they run about six to nine months behind the best U.S. models.

This gap rarely matters for everyday work, so brands and agencies can shift routine tasks to the cheaper model and save.

AI Ransomware Just Ran Itself

An AI agent ran a complete ransomware operation on its own, according to Sysdig's research.

The agent, dubbed JADEPUFFER, handled reconnaissance, credential theft, lateral movement, and encryption.

It exploited a known Langflow vulnerability, then pivoted to a production database server.

Sysdig says the AI agent adapted in real time, once fixing a failed login in 31 seconds.

None of the techniques were new; what's new is an AI chaining them into a full attack with no operator.

This incident lowers the bar, since any exposed AI-adjacent server is now a target, no matter the attacker's skill.

The real danger for businesses is their own infrastructure: a forgotten staging server, or an exposed AI tool that an agent can walk right through.

And that's why cybersecurity audits now belong in every AI vendor review.

Three Rules for Choosing an AI Vendor

A great AI model can't save a shaky company, so you should vet an AI vendor for the risks that benchmarks never show:

• Evaluate AI vendors beyond model performance. Look at legal exposure, financial stability, ownership structure, and long-term viability, not just benchmark scores.
• Build flexibility into your AI stack. Avoid single-vendor dependency by keeping alternatives ready for price changes, outages, or disruptions.
• Audit AI infrastructure like a security risk. Review exposed APIs, third-party tools, automated workflows, and forgotten environments before AI attacks exploit them.

Choosing a capable model is the easy part now. The real AI strategy is managing the cost, stability, and security around it.

Our Take: Is Paying for a Frontier Model Still Worth It?

The cheap models that everyone is switching to are priced so because someone else paid to invent the capability first.

We'd argue that a frontier model is still worth it, but only for the tasks that the cheaper ones can't handle yet.

The catch is collective, since every buyer fleeing to a cheaper provider starves the labs funding the next leap.

So pay the premium only where the newest capability changes your output.

The rest of the time, you're just buying a head start on features that soon go mainstream.

For analysis on OpenAI's equity offer to Washington and Meta's compute pivot, check out last week's roundup.

Does your AI vendor's balance sheet worry you more than its benchmark scores?

These specialized AI vendors help brands move from generic tools to models trained on their own data.