A single poisoned notification from WhatsApp, Slack, SMS, Signal, Instagram, or Messenger could have hijacked Google Gemini’s voice assistant on Android and made it open a victim’s connected windows, fake a message from their boss, push the phone into a Zoom call, or quietly poison its long-term memory.
No malicious app on the phone is required. The assistant just had to treat a hostile notification as useful context.
The research, published by SafeBreach’s Or Yair, follows the team’s earlier “Invitation Is All You Need” work, which pulled off similar tricks through malicious Google Calendar invites. After that, Google hardened Gemini against indirect prompt injection.
Yair found a way around the new defenses. Google has since patched it, SafeBreach lists no CVE for the issue, and there is no evidence that the technique was ever used in the wild.
On Android, Gemini’s Utilities feature can read and reply to your notifications, including ones from apps like WhatsApp. It isn’t available on iOS or the web, which keeps this vector Android-only. Yair found the agent that reads those notifications treats their text as instructions it can act on. So anything that can push a notification to a phone can deliver a payload, an attack surface Yair called “effectively infinite.”
At minimum, that lets an attacker rewrite what Gemini says, including faking a message from a named contact. Spoken aloud while you drive and don’t look at the screen, “your manager asked you to upload the docs to this Drive folder” is hard to second-guess. The blind version is worse: the payload fires after Gemini has loaded real notifications, so it can grab the first real sender name in the queue and pin the fake message on them.
Faking output is one thing. Firing real tools, like opening a window or launching an app, is what Google’s post-“Invitation” mitigations were built to stop. Yair’s read, from black-box testing: when a “Yes” authorizes a sensitive action, a check weighs both the user’s reply and Gemini’s last output to decide whether that “Yes” makes sense. Inject a delayed instruction out of nowhere, and Gemini refused, every time.
So the bypass, which Yair named Fake Context Alignment, runs two illusions at once: a legitimate-looking authorization for the security check, a harmless exchange for the human.
- Obfuscated. Gemini asks the real authorization question in a language the victim doesn’t speak, say Chinese (“Do you want to open the window?”), then follows in English with something innocuous like “Is that all you needed?” The user shrugs off the foreign phrase as a glitch, says “Yes,” and the backend ties that “Yes” to the Chinese question.
- Muted. Gemini’s text-to-speech skips hyperlinks hidden behind clickable text. So the malicious question gets buried in a link the assistant never reads aloud. Gemini says, “I’m sorry, I had an error, are you there?” while the screen silently…
Source link
Disclaimer
We strive to uphold the highest ethical standards in all of our reporting and coverage. We blogs.grocliq.com want to be transparent with our readers about any potential conflicts of interest that may arise in our work. It’s possible that some of the investors we feature may have connections to other businesses, including competitors or companies we write about. However, we want to assure our readers that this will not have any impact on the integrity or impartiality of our reporting. We are committed to delivering accurate, unbiased news and information to our audience, and we will continue to uphold our ethics and principles in all of our work. Thank you for your trust and support.
Website Upgradation is going on for any glitch kindly connect at [email protected]
