Apple Builds an AI Agent as Meta OpenAI and Anthropic Compete

Apple is assembling a team to build its own AI agent, according to several job-listing reports. The planned assistant could browse the web, work across apps, run commands, write and debug code, call services and coordinate specialized subagents to complete multi-step tasks. Apple is seeking a senior engineering leader, with work involving Siri, UI Frameworks and Apple Foundation Models. The listings emphasize privacy, permissions, sandboxing and transparency; Bloomberg has reported a possible launch by the end of 2027. Apple has not commented on the reports.

The effort comes as Apple faces competition from agents and assistants offered by Meta, OpenAI and Anthropic. One report says four of the App Store’s top six free apps are rival AI assistants, including Meta’s Muse at No. 2. Apple reportedly plans to make it harder to grant outside agents access to Macs: a future update would require 14 steps, up from four, to let ChatGPT control a computer, while macOS 27.3 is reported to add warnings. Apple also pays Google about $1 billion annually for Gemini, and CFO Kevan Parekh says the company is still assessing the costs of AI agents. A job listing that described the project later disappeared.

Intuit is positioning its products for a shift in how people discover financial services. CEO Sasan Goodarzi says the company wants to reach customers both on its own platform and through AI platforms, bringing trusted financial data and expertise to AI experiences. In the quarter ended Oct. 31, revenue rose 12% to $4.4 billion and Global Business Solutions revenue increased 15% to $3.1 billion. Intuit agreed to sell Mailchimp to Bending Spoons for about $3.68 billion, with the transaction expected to close in early 2027, subject to approval. Intuit plans to use the proceeds for fiscal 2027 share repurchases; it paid about $12 billion for Mailchimp in 2021, and Mailchimp’s quarterly revenue fell 1% to $313 million.

Intuit lowered its fiscal 2027 revenue outlook to $23.08 billion–$23.37 billion, citing slower expected growth at Credit Karma, while keeping its adjusted earnings forecast at $22.88–$23.12 per share. CFO Sandeep Aujla says AI is changing how customers find financial information, posing both challenges and opportunities for Credit Karma. Separately, KPMG finds that only 1% of insurance CEOs have seen a measurable return on AI investment so far, even as insurance leaders increase spending.

Anthropic’s record IPO has not reversed declines in AI stocks. The company raised $81.34 billion, selling 980 million shares at $83; its shares closed the first trading day at $97.17, up 17%, with demand reported at roughly six times the shares available. Investors are weighing whether AI spending will produce sufficient revenue and returns, alongside interest-rate risks. UBS has lowered its global-equities view to neutral, while Goldman Sachs forecasts the S&P 500 at 8,700 by the end of 2027, a 17% rise. Goldman expects earnings growth to slow from about 30% in 2026 to 18% in 2027 as AI capital-spending growth cools. Senior Democrats have also raised the prospect of tighter rules for technology companies over advanced-AI risks and economic volatility.

AI infrastructure demand continues to stretch supply. Nvidia says demand for its AI systems exceeds what suppliers can provide. Kinsus plans to expand Nvidia package-substrate capacity more than 3.5 times at its Youth campus in Taoyuan, Taiwan, by the third quarter of 2028. Samsung Electro-Mechanics is improving existing operations and upgrading or building facilities in Sejong and Busan, South Korea. Siliconware Precision Industries plans to expand advanced packaging in Taiwan, increasing dedicated Nvidia capacity by mid-2029.

Two research reports highlight risks and limitations in AI systems. Researchers who placed 2 million digital tripwires in requests sent across a large network of AI services found sensitive data leaking through multiple channels; the article says malicious actors were exploiting exposed information. The findings call for stronger safeguards by providers and closer attention to risk by developers using AI services. In a separate study, MM-SystemOneBias evaluated 13 multimodal decision models on 5,323 examples from nine public datasets and 11 task scenarios. Decision models were more stable than matched models under several presentation changes and lost less accuracy when confronted with incorrect external judgments or conflicting descriptions, though answer probabilities sometimes shifted even when the semantic decision stayed the same.

Jacob Coxon, a former Anthropic researcher who left in September after accusing the company of putting profits ahead of safety, told AFP that powerful AI should not be controlled by private companies without adequate oversight. He warned that unchecked growth could cause catastrophic harm, including loss of human life, and called for regulation. The article about Daybreak Blue, meanwhile, is a walkthrough of its onboarding and setup experience and provides no further product details.

Key Takeaways

  • Apple is recruiting for a privacy-focused AI agent that could use apps, browsers, commands, code and services, with a possible launch by the end of 2027 reported by Bloomberg.
  • Apple reportedly plans additional safeguards before outside agents can control a Mac; enabling ChatGPT access may require 14 steps, up from four.
  • Meta, OpenAI and Anthropic are among Apple’s AI competitors; Apple also pays Google about $1 billion a year for Gemini.
  • Intuit reported quarterly revenue of $4.4 billion, up 12%, and agreed to sell Mailchimp to Bending Spoons for about $3.68 billion.
  • Intuit lowered its fiscal 2027 revenue forecast to $23.08 billion–$23.37 billion but kept its adjusted earnings outlook unchanged.
  • Anthropic raised $81.34 billion in its record IPO; shares priced at $83 and closed their first day at $97.17, while AI stocks continued to slide.
  • Nvidia suppliers are expanding capacity: Kinsus targets more than 3.5 times its package-substrate capacity by Q3 2028, while SPIL plans additional dedicated capacity by mid-2029.
  • KPMG says just 1% of insurance CEOs have recorded measurable returns from AI investment, despite plans to increase spending.
  • Researchers found sensitive data leaking across AI services after testing them with 2 million digital tripwires.
  • A study of 13 multimodal decision models found greater stability under several presentation changes, but shifts in answer probabilities could still occur without a change in semantic decisions.

Apple appears to be building a private AI agent for Mac

Apple appears to be developing an AI agent that can use a computer to complete tasks for users. A job listing says it could browse the web, work across apps, run commands, write and debug code, and use services. The listing says the agent would handle multi-step tasks while moving between tools. Apple did not respond to TechCrunch’s request for comment. Bloomberg also reported that macOS 27.3 will show warnings when users give third-party agents full access to their computers.

Apple seeks engineering leader for privacy-focused AI agent

Apple is seeking a senior engineering leader to help build an AI agent, according to a job listing published on November 19, 2026. The listing describes an agent that can browse the web, work across apps, run commands, write and debug code, and call services. It says the agent should choose the right tool for each task and work privately. The team is building the foundation for this system.

Apple reportedly plans a full-featured AI agent for users

A job listing describes Apple’s plan to build an AI agent that can browse the internet, use apps, write code, run commands, and complete tasks across devices and services. Bloomberg reported that the listing seeks a senior engineering manager for a proactive assistant, and mentioned a possible launch by the end of 2027. The listing says the agent should coordinate specialized subagents and include strong permissions, sandboxing, and transparency. Apple’s rivals, including Meta, OpenAI, and Anthropic, already offer AI agents. Bloomberg also reported that Apple is adding warnings and extra steps before users give outside agents computer access on Mac.

Apple’s AI agent plans signal a push to catch up

An Apple job listing described an AI agent that could browse the web, work across apps, and write code, but the listing later disappeared. The report said CEO John Ternus, who replaced Tim Cook on September 1, has urged software chief Craig Federighi to move faster on AI. AI assistants from Apple’s rivals hold four of the App Store’s top six free app spots, with Meta’s Muse at No. 2. Bloomberg reported that Apple’s next Mac update will require 14 steps, up from four, to let ChatGPT control a computer. Apple also pays Google about $1 billion a year for Gemini, and CFO Kevan Parekh said the company is still in the early stages of understanding the costs of AI agents.

Apple hires engineers to develop its own AI agent

Apple is working on an AI agent that could browse the web, work across apps, write code, and call services, according to a job listing. The listing says the agents will plan and carry out multi-step tasks across apps, browsers, command lines, code, and APIs. They could also coordinate specialized subagents. The engineer would work with teams in Siri, UI Frameworks, and Apple Foundation Models.

Intuit CEO sees AI tools as new paths to customers

Intuit CEO Sasan Goodarzi said AI tools and agents are becoming new ways for customers to find information and services. Intuit aims to reach customers on its own platform and through AI platforms, working with leading AI companies to build new experiences. Goodarzi said Intuit brings trusted financial data and expertise to help people take financial actions safely and accurately. In the quarter ended Oct. 31, revenue rose 12% to $4.4 billion, while Global Business Solutions revenue grew 15% to $3.1 billion. Intuit also agreed to sell Mailchimp to Bending Spoons, saying the move will sharpen its focus on AI-based services.

Intuit plans to sell Mailchimp for $3.68 billion

Intuit plans to sell Mailchimp, its email marketing platform, to Bending Spoons for about $3.68 billion. The deal is expected to close in early 2027, subject to approval, and Intuit plans to use the proceeds for fiscal 2027 share repurchases. Intuit paid about $12 billion for Mailchimp in 2021, and the business’s quarterly revenue fell 1% to $313 million. Intuit lowered its fiscal 2027 revenue forecast to $23.08 billion to $23.37 billion because it expects slower growth at Credit Karma, but it kept its adjusted earnings forecast of $22.88 to $23.12 a share. CFO Sandeep Aujla said AI is changing how customers find financial information, creating challenges and opportunities for Credit Karma.

AI stocks slide after Anthropic’s record IPO as investors debate what comes next

AI stocks continued to fall after Anthropic’s record IPO failed to calm markets. Anthropic priced shares at $83 and closed its first day of trading at $97.17, up 17%, after raising $81.34 billion. Wall Street opinions differ: UBS lowered its view of global equities to neutral, while Goldman Sachs predicts the S&P 500 will rise 17% to 8,700 by the end of 2027. Investors are watching whether AI spending can produce enough revenue and returns, as well as the effect of interest rates. Goldman expects earnings growth to slow from about 30% in 2026 to 18% in 2027 as growth in AI capital spending cools.

Former Anthropic researcher warns about AI power in private companies’ hands

Jacob Coxon, a former researcher at Anthropic, told AFP that it is “insane” for powerful AI technology to be controlled by private companies. Coxon left Anthropic in September after accusing the company of putting profits ahead of safety. He said the AI industry lacks enough regulation and oversight and warned that unchecked growth could lead to catastrophic harm, including loss of human life. He called for regulation to prevent what he described as a disaster.

KPMG finds only 1% of insurance CEOs see returns from AI so far

KPMG reports that just 1% of insurance CEOs have seen a measurable return from their investment in artificial intelligence so far. Even so, insurance leaders are increasing their investment in the technology.

Nvidia expands supplier capacity to keep up with AI demand

Nvidia suppliers Kinsus Interconnect Technology, Samsung Electro-Mechanics and Siliconware Precision Industries agreed to expand production for Nvidia’s AI computing platforms. Kinsus plans to increase package substrate capacity for Nvidia more than 3.5 times at its Youth campus in Taoyuan, Taiwan, by the third quarter of 2028. Samsung Electro-Mechanics will improve operations at existing factories and is upgrading facilities and building factories in Sejong and Busan, South Korea. SPIL plans to expand advanced packaging in Taiwan to increase Nvidia’s dedicated capacity by mid-2029. Nvidia says demand for AI systems continues to exceed supply.

Researchers find sensitive data leaks across AI services

Researchers used 2 million digital tripwires in requests sent through a large network of AI services to test whether sensitive information could leak. They found that many services did not properly protect user data and that information leaked through different channels. The article says malicious actors were exploiting the exposed data. The findings point to the need for developers to understand the risks of using AI services and for providers to strengthen security measures.

See how to get started with Daybreak Blue

A walkthrough video shows how to onboard to Daybreak Blue. It gives a closer look at the setup experience and explains how to get started.

Study tests bias in multimodal decision models

Researchers introduced MM-SystemOneBias to test how multimodal decision models respond to changes in how information and choices are presented. The study evaluated 13 models on 5,323 examples from nine public datasets and 11 task scenarios, using six types of changes. Across seven matched model and backbone pairs, decision models were more stable under changes to option labels, irrelevant information and rating scales. They also had smaller accuracy losses when faced with incorrect external judgments in all seven pairs and conflicting descriptions in six. However, the researchers found shifts in answer probabilities even when the models kept the same semantic decision.

Anthropic’s record IPO fails to lift falling AI stocks

Anthropic’s IPO raised $81 billion through the sale of 980 million shares, making it the largest IPO in history, according to the article. Demand for the shares was about six times the number available. The debut came as prices for other AI stocks continued to fall, despite the sector’s major role in US economic growth and investment in data centers powering Claude and ChatGPT. Senior Democrats have suggested tech companies could face tighter rules over risks from advanced AI and the economic volatility that AI IPOs may bring.

Sources

NOTE:

This news brief was generated using AI technology (including, but not limited to, Google Gemini API, Llama, Grok, and Mistral) from aggregated news articles, with minimal to no human editing/review. It is provided for informational purposes only and may contain inaccuracies or biases. This is not financial, investment, or professional advice. If you have any questions or concerns, please verify all information with the linked original articles in the Sources section below.

Apple AI agent AI assistants Siri Apple Foundation Models AI agent privacy macOS security Intuit Credit Karma Mailchimp AI investment returns Anthropic IPO AI stocks AI infrastructure Nvidia advanced packaging AI data leakage multimodal AI AI model robustness AI regulation AI safety

Comments

Loading...