OpenAI has launched an Apple iMessages plug-in for ChatGPT that allows users to connect their Messages inbox to the chatbot, giving the AI system access to conversations for tasks ranging from searching message histories to drafting and sending texts.

The feature expands ChatGPT’s role from a standalone chatbot into an assistant capable of interacting with users’ personal communications. OpenAI said the plug-in can sort, analyse, and edit messages, while users can also ask ChatGPT to find information contained in older conversations.

The integration works with Codex and ChatGPT Work, meaning the functionality can also be used in professional settings rather than being limited to personal messaging.

A promotional video for the feature shows a user asking ChatGPT to suggest follow-up messages to contacts based on conversations received the previous day. The tool can also be instructed to delete messages, draft responses, and send messages on a user’s behalf.

That level of access makes the integration one of the more consequential extensions of AI assistants into everyday digital activity. Instead of simply generating text in response to a prompt, ChatGPT can use a user’s existing conversations as context and take actions inside the messaging environment.

The development also raises questions about how much personal data an AI assistant needs to access to perform those tasks and how that information is processed.

OpenAI told Bloomberg that the plug-in operates locally on the user’s machine and “doesn’t create an index of all someone’s messages.” The company has not publicly provided all of the technical details needed to explain how message data is accessed, processed, and retained under the integration.

The ability to send messages introduces another layer of risk because an AI system is not only interpreting private conversations but can potentially act on a user’s behalf.

OpenAI advises users to monitor ChatGPT’s activity and discourages enabling persistent approval for message-sending actions. The company warns that persistent approval “removes your final chance to review a message before ChatGPT sends it as you.”

That warning underpins the distinction between using AI as a drafting tool and allowing an agent to execute actions autonomously. A generated response can be reviewed and changed before it is sent, while autonomous sending creates the possibility that an inaccurate interpretation, inappropriate tone, or mistaken recipient could result in a message being delivered without a final human check.

The Messages integration is part of a broader push by OpenAI to make ChatGPT an agent capable of interacting with applications and data on a user’s device. The company’s strategy involves connecting its models to external tools so they can search information, manipulate files, write code, and perform tasks rather than simply answer questions.

For users, the appeal is convenience. Conversations that once required manually searching through years of messages can potentially be retrieved with a natural-language request. Routine follow-ups can be drafted from existing context, while professional users could use the same capabilities to work with communications without repeatedly switching between applications.

The trade-off is that the more deeply an AI assistant is integrated into personal and professional workflows, the greater the importance of clear controls over access, permissions, and actions.

OpenAI’s decision to keep the processing local, as described by the company, is intended to address some of those concerns. But the practical privacy implications will depend on how the plug-in communicates with ChatGPT, what information leaves the device, how permissions are enforced, and whether users can easily determine what the system has accessed.

The Messages integration therefore marks a significant expansion of ChatGPT’s capabilities, but also puts greater emphasis on the safeguards surrounding agentic AI. The technology is moving from helping users write messages to potentially reading, organizing, and sending them. But safety advocates warn that this makes human oversight and transparent permission controls necessary as AI assistants gain access to more personal data.

CNBC’s Jim Cramer said Thursday that rising bond yields, higher oil prices and concerns about the U.S. consumer are making it increasingly difficult for investors to reward companies with strong underlying businesses, creating what he described as an “incredibly jarring gulf between stock prices and reality.”

Cramer’s comments came after another difficult session for Wall Street, with the Dow Jones Industrial Average falling 1.3%, the S&P 500 declining nearly 0.9% and the Nasdaq Composite losing 1%.

The selloff reflected growing concerns that higher energy prices, driven by the conflict involving Iran, could keep inflation elevated and limit the Federal Reserve’s ability to ease monetary policy. Treasury yields also moved higher, reversing much of the decline that followed the Treasury Department’s announcement Wednesday of a plan to bring down longer-term borrowing costs.

The 30-year Treasury yield had climbed above 5.33% earlier in the week, reaching a level not seen in nearly two decades.

Cramer argued that the market environment is forcing investors to evaluate individual companies through a much broader macroeconomic lens, even when their businesses continue to perform well.

“There’s an incredibly jarring gulf between stock prices and reality,” Cramer said on “Mad Money.”

He made the comments from the construction site of Micron Technology’s massive semiconductor fabrication complex in Boise, Idaho, where thousands of workers are building facilities that will eventually manufacture memory chips used in artificial intelligence systems.

The project, Cramer said, illustrates the scale of investment taking place in U.S. manufacturing and AI infrastructure at a time when financial markets are increasingly focused on economic risks.

“Unfortunately, though, you can’t take your eye off the broader market even if you think, as I do, that Micron’s stock is radically undervalued,” he said. “In the end, we always have to look at stocks through the market’s prism.”

Micron shares rose about 4% Thursday, even though the stock remains roughly 20% below its June record high. Cramer’s Charitable Trust, the portfolio associated with CNBC’s Investing Club, owns Micron shares.

The performance stood in contrast to the broader market, where concerns about consumers, energy costs, and interest rates continued to dominate trading.

Walmart provided one of the clearest examples of how those pressures can affect a major company. Shares of the retail giant plunged about 9% Thursday after the company reported quarterly comparable sales below Wall Street expectations and issued sales guidance that also disappointed investors.

Cramer said the results were more complicated than the headline figures suggested. Walmart has continued to emphasize low prices and market-share gains rather than maximizing short-term profit, while higher gasoline prices have reduced the amount of money consumers have available for other purchases.

Gasoline prices above $4 a gallon could place additional pressure on household budgets, particularly if elevated oil prices persist as the conflict involving Iran continues. That creates a difficult backdrop for retailers and other consumer-facing businesses because higher fuel costs can simultaneously increase operating expenses and reduce consumers’ discretionary spending.

“Two-thirds of this country’s economy is service-based,” Cramer said, explaining that the strength of the American consumer remains more important to the broader economy than the manufacturing investment taking place at projects such as Micron’s Idaho facility.

The bond market represents another challenge.

Treasury Secretary Scott Bessent told CNBC Thursday that the Treasury’s planned purchases of longer-dated government debt could exceed the $4 billion upper limit discussed the previous day.

The proposed purchases are intended to help put downward pressure on longer-term Treasury yields, but Cramer questioned whether the intervention would be large enough to make a meaningful difference given the size of the U.S. government’s debt.

“When America has $40 trillion in debt, a $4 billion buyback has the Treasury Secretary looking like the Little Dutch boy with his finger plugging the dike,” Cramer said.

The comparison underscores the scale mismatch between the Treasury’s proposed intervention and the broader forces influencing the bond market. Investors are weighing government borrowing needs, inflation, monetary policy and geopolitical risks, all of which can exert upward pressure on long-term yields.

Higher yields have much bearing for equities because they increase the return investors can obtain from relatively lower-risk government securities while also raising the discount rate applied to future corporate earnings. That can place disproportionate pressure on growth and technology stocks whose valuations depend heavily on profits expected years into the future.

Cramer said Micron’s performance demonstrates the difficulty of separating individual corporate fundamentals from broader market sentiment. The company is benefiting from substantial investment in AI infrastructure and from efforts to expand U.S. semiconductor manufacturing, yet its valuation remains affected by changes in interest rates and overall risk appetite.

“Micron’s stock finished up 4%. That’s terrific American exceptionalism at work,” Cramer said. “The problem is there are another 499 stocks in the S&P 500 and the prism made a lot of them look downright awful today.”

This points to a growing divide within the U.S. stock market. Companies tied to AI, semiconductor manufacturing, and other areas of strategic investment can continue to experience strong underlying demand, while their shares remain vulnerable to macroeconomic shocks.

OpenAI is regaining ground on Anthropic among U.S. businesses, new data from corporate spending platform Ramp shows, offering an early indication that competition between the two leading AI companies is becoming increasingly fluid as enterprises experiment with competing models.

Neither OpenAI nor Anthropic has publicly disclosed the detailed financial information investors will eventually expect to see as the companies move closer to potential initial public offerings. In the meantime, corporate spending data can provide an imperfect but useful window into how businesses are allocating money across AI providers.

Ramp, which provides corporate cards, bill payment and expense management services, tracks spending patterns across more than 70,000 U.S. businesses. Its customers range across industries, although the company’s concentration in technology startups and other venture-backed companies means the data is not representative of the entire American corporate economy.

The latest figures show Anthropic maintaining its lead among Ramp’s paying business customers, but OpenAI is beginning to close the gap.

Anthropic accounted for nearly 44% of spending among the two companies in July, compared with nearly 40% for OpenAI, according to Ramp. The figures measure the share of Ramp’s business customers paying for products from the two AI companies, rather than total revenue or the amount of money spent.

The shift began in May, when Anthropic overtook OpenAI for the first time among Ramp’s paying business users. Anthropic reached 41% at the time, compared with OpenAI’s 39%.

OpenAI has not reclaimed the top position since then. But Ramp economist Ara Kharazian said OpenAI was growing faster among the segment during the third quarter so far, suggesting the gap could narrow again.

“GPT-5.6 Sol is really good, increasingly the choice for developers,” Kharazian said in a post on X, attributing part of OpenAI’s recent momentum to its latest model.

The data provides a useful counterpoint to the idea that Anthropic has established a durable lead in enterprise AI. Anthropic’s rise among businesses has been one of the most closely watched developments in the AI market. Its Claude models have developed a strong following among software developers and companies seeking AI systems for coding, research, and other professional applications.

OpenAI, meanwhile, has historically benefited from ChatGPT’s enormous consumer user base and broad enterprise adoption. The company’s challenge has been converting that early lead into sustained business spending as rivals improve their models and target specific professional workflows.

Ramp’s numbers suggest that enterprise customers remain willing to switch between providers as new models emerge. That creates an important question for investors: how “sticky” is enterprise AI spending?

Traditional enterprise software tends to become deeply embedded in company workflows, creating switching costs that can make customers reluctant to move to competing products. AI may prove different because companies can test several models simultaneously, route different tasks to different systems, and change providers when a new model offers better performance, lower prices, or more favorable terms.

The result could be a more volatile enterprise software market in which model releases have a direct and immediate impact on corporate purchasing decisions.

Anthropic’s lead also needs to be interpreted carefully. Ramp does not disclose the actual dollar value of spending represented by the percentages, and its dataset excludes companies that use competing corporate-spending platforms, including large businesses that manage expenses through providers such as American Express.

That makes Ramp’s figures an indicator of market direction rather than a comprehensive measure of OpenAI or Anthropic’s enterprise revenue.

The composition of Ramp’s customer base also matters. Its concentration among technology companies and startups could make its customers more likely than the broader corporate market to experiment with multiple AI models, adopt new developer tools, and rapidly shift spending following major model launches.

Even with those limitations, the data points to a broader trend that could be more important than the competition between OpenAI and Anthropic themselves: corporate adoption of paid AI services is continuing to expand.

Among Ramp’s customers, the percentage of businesses paying for AI products rose to nearly 56% in July, from more than 50% in March. That means OpenAI and Anthropic can both increase their business revenue even while competing for the same customers and losing relative market share to each other.

Market-share gains do not necessarily mean one company is taking revenue directly from another. If the number of businesses purchasing AI products continues to rise, both providers can expand rapidly while their relative positions fluctuate.

The model race is making that competition even more dynamic.

Companies are now evaluating AI systems based on coding performance, reasoning ability, agentic capabilities, price, latency, security, data controls, and integration with existing software. A model that wins on one of those dimensions can gain adoption quickly, while a rival can recover ground with its next release.

Anthropic’s Fable 5, according to Kharazian, had weaker adoption in Ramp’s data, which he attributed partly to its pricing and regulatory-related data-retention requirements. Anthropic has faced user concerns over its policy requiring Fable users to retain data for 30 days in certain circumstances.

Still, attributing changes in market share to a single model release would be premature. Enterprise purchasing decisions are influenced by a combination of model performance, pricing, procurement policies, security requirements, existing contracts, and how easily a system can be incorporated into a company’s workflows.

The larger takeaway is that the U.S. enterprise AI market is entering a more competitive phase.

The first stage of generative AI adoption was dominated by experimentation, with companies testing ChatGPT and competing systems to determine where the technology could create value. The market is now moving toward a phase in which businesses are paying for AI at scale and evaluating competing models more systematically.

Ramp’s data indicates that this transition is benefiting the market as a whole. More than half of the company’s tracked businesses now pay for AI, and that proportion continues to rise. That expansion could matter for OpenAI and Anthropic as both companies approach a stage where investors will demand greater visibility into their financial performance.