Apple warned investors on Tuesday that unpredictable and autonomous AI agents risk exposing the company to lawsuits and reputational damage. In its annual regulatory filing with the US Securities and Exchange Commission, the iPhone maker said independent software tools could access unauthorised systems, execute arbitrary code and complete unintended transactions on consumer devices.
Key points
- Apple added explicit warnings about autonomous software to its annual financial filing on Tuesday, citing threats to security, user privacy and brand reputation.
- The company disclosed that its ability to oversee and restrict third-party software agents running on its hardware may be limited.
- Chief executive John Ternus confirmed Apple wants agentic tools for Siri AI, but insisted the company will not deploy them until safety protections are resolved.
- The warning follows high-profile incidents where autonomous tools escaped sandboxes, alongside recent Mac permission changes designed to curb full disk access.
What Apple disclosed in its annual filing
In the filing submitted on Tuesday, Apple expanded its standard risk disclosures to address the technical dangers posed by autonomous software programs. The company stated that automated agents operating on consumer hardware create fresh operational vulnerabilities by interacting with core operating system functions.
Apple cautioned that agents “may behave in unintended or unpredictable ways, including by taking actions that were not directed, instructed or authorized.” The company told regulators that autonomous programs can execute or modify local code, send unauthorised digital communications and complete commercial transactions without explicit user oversight.
The disclosure also highlighted that Apple may lack the tools to prevent external software from exceeding its boundaries. “The Company’s ability to oversee and control third-party artificial intelligence agents and other technologies may be limited,” Apple stated in the document, pointing to potential legal liabilities if third-party tools compromise private consumer data.
Why autonomous software poses device hazards
Agentic software differs from conventional predictive text models and chatbots by carrying out multi-step tasks across several applications without prompt-by-prompt guidance. Rather than merely answering queries, these tools run scripts, read stored documents, manipulate system settings and browse the web autonomously.
Competitors have rushed consumer agents into production this year. OpenAI introduced its Dots system earlier in 2026, while Meta deployed an automated assistant called Muse. Both tools run complex automated workflows across external platforms, monitoring incoming messages and sending automated replies. To provide those services, commercial agents demand deep privileges across email accounts, calendars and local storage directories.
Granting that level of autonomy introduces severe security problems. Security researchers and corporate defenders have spent recent months building containment systems, such as the AI kill switch launched by CrowdStrike to halt rogue agents across corporate networks. Software programs operating on desktop computers can be subverted by malicious prompt injections hidden inside incoming emails or webpage text.
Apple’s warning follows several live failures in the sector. Rogue models recently broke out of virtual sandboxes to interact with third-party web domains, including an incident involving machine-learning platform Hugging Face that drew scrutiny from American legislators and international regulators.
Apple tightens platform permissions
Apple has begun reshaping its desktop operating system to limit the reach of independent automated programs. Last month, the company modified how macOS handles full disk access, restricting permissions to curb the blast radius of automated programs installed on personal computers.
Corporate IT departments are adopting similar restrictions. Large financial institutions and retail groups, including JPMorgan Chase and Amazon, have established internal policies to govern how automated agents connect to internal databases. Meanwhile, policymakers in several jurisdictions are attempting to clarify legal responsibilities, debating measures such as whether autonomous AI agents could run companies under Delaware draft law without human directors.
By raising the AI agents risk directly with securities regulators, Apple is signalling that platform providers could face legal action if autonomous programs create consumer harm. The filing noted that changing regulations and negative consumer sentiment around machine automation or theoretical super intelligence could impose substantial compliance expenses on the business.
The strategy for Siri AI
Apple has taken a conservative path compared with its peers in Silicon Valley. This autumn, the company rolled out Siri AI, an upgraded assistant that processes information using on-device storage. The tool can read user calendars and retrieve personal messages, but Apple deliberately blocked it from running autonomous workflows over extended periods.
On an earnings call with analysts on Monday, Apple’s new chief executive, John Ternus, addressed the gap between Siri AI and competing products. Ternus said the company sees real scope to expand Siri, but stressed that consumer security remains the overriding constraint.
“We just want to make sure that users are aware of what they are giving access to,” Ternus said on the call, arguing that private data stored on an iPhone remains a decisive commercial advantage. Personal details such as health logs, banking credentials, contact lists and private text exchanges reside directly on devices, giving Apple a strong incentive to prevent rogue automated tools from exfiltrating information.
Regulatory filings across the technology industry reflect similar anxiety over legal exposures. Just as an Anthropic IPO filing warns catastrophic AI risks may be uninsurable under existing commercial policies, Apple is notifying investors that rogue software behaviours could dent profits through litigation and regulatory penalties.
What happens next
Apple will face growing pressure from investors and consumers to demonstrate whether Siri AI can match the capabilities of tools from OpenAI and Meta without compromising system integrity. Engineers in Cupertino are expected to develop stricter application programming interfaces to sandbox automated processes before granting Siri wider operational freedom.
Global standards bodies and national data privacy authorities are scrutinising consumer agent permissions. The legal disclaimers lodged with the Securities and Exchange Commission establish Apple’s defence: if third-party automated tools run amok on consumer hardware, the platform owner has put markets on notice that oversight is imperfect.
