Apple Intelligence Faces 76% Success Rate in Prompt Injection Attacks
Severity: High (Score: 67.5)
Sources: Foxnews
Published: · Updated:
Keywords: apple, your, privacy, intelligence, really, spent, years
Summary
Research from RSAC Research reveals vulnerabilities in Apple's on-device AI, Apple Intelligence, allowing for prompt injection attacks with a 76% success rate. The researchers disclosed their findings to Apple on October 15, 2025, leading to security enhancements in iOS 26.4 and macOS 26.4. These attacks can manipulate the AI without needing physical access to the device, relying instead on cleverly crafted text. The techniques used include Neural Exec and Unicode's right-to-left override feature, which can hide malicious instructions. This poses risks as attackers can potentially manipulate sensitive data or app functions. Users are advised to understand these risks and adjust their privacy settings accordingly. Key Points: • Apple Intelligence is vulnerable to prompt injection attacks with a 76% success rate. • The attack methods include Neural Exec and Unicode manipulation techniques. • Apple has released patches in iOS 26.4 and macOS 26.4 to address these vulnerabilities.
Detailed Analysis
**Impact** Apple users globally who rely on Apple Intelligence for on-device AI tasks are affected, including iPhone, iPad, and Mac users. The attack could lead to unauthorized manipulation of AI responses, potentially impacting apps with access to sensitive data or system functions. This creates risks for personal data confidentiality and app integrity without requiring physical device access or server breaches. No specific sectors or geographic concentrations were detailed. **Technical Details** The attack uses prompt injection via adversarial prompts and Unicode right-to-left override characters to bypass AI safeguards. Two main techniques identified are Neural Exec, which confuses the model with malformed prompts, and Unicode tricks that hide malicious instructions. The attack targets the on-device large language model powering Apple Intelligence, exploiting the AI’s interpretation of hidden commands during text processing. No CVEs or malware names were provided. **Recommended Response** Apply the security updates included in iOS 26.4 and macOS 26.4, which Apple released to harden protections against these prompt injection attacks. Review app permissions and limit Apple Intelligence access where possible. Monitor for suspicious input containing unusual Unicode characters or malformed prompts in apps interfacing with Apple Intelligence. No additional IOCs or detection signatures were specified.
Source articles (2)
- Apple Intelligence vulnerable to prompt injection with 76% success rate — Foxnews · 2026-06-07
Apple has spent years telling us that privacy starts on the device. For many users, that message feels reassuring. Your messages, photos, emails and app data sit in your hand, protected by Face ID, pa… - Is Apple Intelligence on your iPhone really secure? — Foxnews · 2026-06-07
Tech expert Kurt Knutsson urges you to use Apple’s App Privacy Report to see what your apps are really up to. Apple has spent years telling us that privacy starts on the device. For many users, that m…
Timeline
- 2025-10-15 — RSAC Research disclosed findings to Apple: Researchers informed Apple about the vulnerabilities in Apple Intelligence, prompting security updates.
- 2026-06-07 — Foxnews publishes vulnerability report: Foxnews reports on the 76% success rate of prompt injection attacks against Apple Intelligence.
Related entities
- Phishing (Attack Type)
- cyberguy.com (Domain)
- tools.it (Domain)
- T1566 - Phishing (Mitre Attack)
- Android (Platform)
- Apple Intelligence (Platform)
- IOS (Platform)
- IPad (Platform)
- IPhone (Platform)
- Mac (Platform)
- MacOS (Platform)
- Private Cloud Compute (Platform)
- Windows (Platform)