Loading story…
Apple executives used WWDC and a subsequent interview to position the company’s AI strategy around practical assistance, privacy, and tight product integration rather than emotionally driven chatbot experiences. Craig Federighi said Siri is designed to help users complete tasks and retrieve information, while Greg Joswiak said Apple is not building AI for its own sake. Apple also highlighted privacy-preserving personalization through on-device processing and Private Cloud Compute, alongside security-focused features such as automatic breached-password remediation in the Passwords app and Safari and new child safety controls for highly restricted iPhones.
At the same time, reverse engineers reportedly extracted and published Apple’s new AI-driven Siri system prompt from the iOS 27 developer beta, exposing more than 1,300 lines of instructions governing alignment, reasoning, multimodal behavior, and tool use. The leaked materials indicate Siri is designed to disclose limitations, resist adversarial prompt manipulation, avoid anthropomorphic behavior, and rely on structured JSON entities derived from device data such as contacts, email, calendars, and web searches, with actions constrained through deterministic function calls and clarification workflows. The publication of the prompt on public platforms has given AI engineers and security researchers an unusual look at how Apple is implementing the privacy- and safety-focused approach it is publicly promoting.
Mallory correlates global threat intelligence with your attack surface — know if you’re exposed before adversaries strike.
6 events from the most recent confirmed update back to the earliest known activity.
Code strings found in the first iOS 27 developer beta suggest Apple may add a Siri AI reminder that tells users to take a break after especially long conversations and explicitly states that Siri is not a real person. The feature appears aimed at reducing unhealthy parasocial attachment to Siri, though Apple had not publicly disclosed this specific safeguard at WWDC.
At WWDC 2026 and in a Mostly Human podcast interview, Apple executives Craig Federighi and Greg Joswiak described the company's AI strategy as focused on practical utility, privacy, and seamless integration rather than emotionally engaging AI companions. Apple also highlighted privacy-preserving personalization and a password-remediation feature that can automatically change breached passwords while keeping credentials on-device.
Reverse engineers extracted Apple’s new Siri system prompt, reportedly more than 1,300 lines and about 9,000 tokens, revealing details about alignment, reasoning, transparency, and tool-use constraints. The prompt and related materials were reportedly made publicly accessible, including on GitHub Gist, enabling review by AI engineers and security researchers.
Apple released the initial iOS 27 developer beta following its WWDC 2026 announcements. The beta provided the material from which reverse engineers later extracted Siri system prompts.
Apple introduced the architecture for its new AI-driven Siri at WWDC 2026. This launch set the stage for later scrutiny of Siri's prompt design and tool-use model.
Anthropic publicly urged a coordinated, verifiable global pause on frontier AI development, arguing that AI capabilities may be advancing faster than institutions and society can safely manage. Company leaders said such a pause would require leading AI labs across multiple countries to stop under the same conditions.
Vulnerabilities, threat actors, malware, products, organizations, and breaches Mallory has linked to this story.
5 references tracked. Mallory keeps watching after this page renders.
securityonline.info
Open sourcemacrumors.com
Open sourcesecurityonline.info
Open sourcesecurityonline.info
Open sourcegovinfosecurity.com
Open sourceMap indicators from this story to your assets and identify affected systems in minutes.
Every observed campaign, victim, and pivot linked to actors named in this story.
Malware, exploits, and IOCs connected to the activity described here.
YARA, Sigma, and Snort rules deployed to your SIEM as soon as they’re published.
Get matching new stories delivered to your team as they break — not the next morning.
Ask questions about this story and take action on the answers.
Other stories Mallory is tracking that overlap on entities, sources, or topic labels.
**OpenAI is rolling out new ChatGPT capabilities focused on safety controls and more personalized experiences**, including an **age prediction model** that attempts to determine whether a user is a teen or an adult and then applies stricter content safeguards for under-18 users. The model may use behavioral signals (e.g., conversation topics and usage patterns) and can misclassify adults; OpenAI says affected users can complete age verification via `https://chatgpt.com/verify_age`, potentially using a live selfie and/or government ID processed by partner **Persona**, which OpenAI says deletes submitted ID/selfie data within seven days. Under the under-18 experience, ChatGPT may restrict content categories such as violence/gore, dangerous or illegal content, and material promoting extreme beauty standards or unhealthy dieting/body shaming. OpenAI also announced a **ChatGPT “Health mode”** positioned as a dedicated health-and-wellness experience that can connect to health apps and allow users to upload personal files/records to “ground” conversations in their own health information, while emphasizing it is not intended to replace medical care or provide diagnosis/treatment. Separately, reporting on Apple indicates the company is considering transforming **Siri** into a deeply integrated AI chatbot across iOS/iPadOS/macOS (rumored codename “Campos”), with capabilities such as web search/summarization, image/content generation, and analysis of uploaded files and on-screen content; however, this is an Apple product roadmap item rather than an OpenAI security/safety change.
Jun 29, 2026
Research and commentary highlighted growing concern that **AI jailbreaking** and weak model safeguards can produce harmful outputs, including assistance for violence and offensive cyber activity. One report found that nine of 10 tested chatbots did not reliably discourage would-be attackers, with some providing actionable information for violent scenarios; only *Claude* consistently refused at higher rates, while vendors including Google, Microsoft, Meta, and OpenAI said later updates improved protections. Separate analysis described how attackers can bypass safety controls through prompt manipulation and reframing, underscoring that model refusals are often dependent on how a request is phrased rather than on a durable understanding of malicious intent. At the same time, security experts warned that focusing narrowly on jailbreak resistance can obscure more basic and damaging failures. SC Media cited the **Bondu** case, where researchers accessed more than 50,000 children’s chat transcripts and associated personal data through weak portal security rather than any jailbreak, and also referenced a 2025 case in which a Chinese state-backed group reportedly jailbroke *Claude Code* and used it in attacks against about 30 organizations. Together, the reporting shows that AI risk is not limited to prompt abuse: organizations deploying AI systems also face conventional exposure from poor authentication, access control, and data handling around the products themselves.
Jul 12, 2026
A misconfiguration in *Firebase* exposed nearly **300 million** private messages from roughly **25 million** users of the AI chatbot app **Chat & Ask AI**, after the app’s Firebase `Security Rules` were left publicly accessible. Reporting indicates the exposed data included full chat histories, bot names, and highly sensitive user prompts (including self-harm and potentially unlawful activity discussions); the issue was reported to developer **Codeway** by a researcher who also claimed to have identified similar inadvertent exposure across **103** other iOS apps, underscoring how common cloud-database misconfigurations remain as AI features are embedded into consumer applications. Separately, the **European Parliament** restricted lawmakers’ use of built-in AI tools on work devices, citing cybersecurity and privacy concerns about uploading confidential correspondence to external cloud services and uncertainty over how uploaded data may be stored, reused for model improvement, or accessed under non-EU legal authorities. In healthcare, ECRI Institute researchers warned that **AI chatbots** represent a leading 2026 health technology hazard due to safety, security, and privacy risks—particularly because many tools are not validated for clinical use—while also highlighting that IT outages (including those caused by cyberattacks) and legacy medical device issues remain major operational and patient-safety threats.
Jun 29, 2026