Perplexity Comet Artificial Intelligence browser vulnerable to prompt injection exploit

Brave published details of a vulnerability in Perplexity’s Comet Artificial Intelligence browser that can be triggered when the browser summarizes a page, allowing injected prompts to access data in other open tabs.

Brave published details about a security issue affecting Comet, Perplexity’s Artificial Intelligence browser, that could allow an attacker to inject a prompt into the browser and access data held in other open browser tabs. The article was published on August 26, 2025 and is credited to Search Engine Journal. The reporting links to Brave’s blog post for the technical disclosure and includes a source link to Search Engine Journal’s coverage.

According to the coverage, the vulnerability can be activated when a user asks the Comet Artificial Intelligence browser to summarize a web page. In that workflow the large language model will read the web page and process content on the page. Brave’s writeup says the model will also process any embedded prompts present on the page that command the model to take actions affecting open tabs. Brave’s account explains how those embedded instructions can be treated as input to the model, creating a prompt injection vector that was sufficient to extract or expose data from other tabs.

The article quotes Brave’s explanation of the vulnerability but the quoted passage is truncated in this report. Readers seeking full technical details, example exploit chains, and Brave’s mitigation advice are directed to Brave’s original blog post, which is linked in the coverage. Beyond the summary of Brave’s disclosure provided here, the report does not state whether Perplexity has released a patch or what immediate mitigations are available to users; the linked Brave post contains the disclosure material referenced by Search Engine Journal.

70

Impact Score

Governance gaps emerge as agentic Artificial Intelligence scales

Agentic Artificial Intelligence is moving from assisted chatbots to autonomous workflows faster than enterprise governance is adapting. The shift raises accountability, security, lifecycle, and cost control challenges that organizations must address in operational code from the start.

Where OpenAI technology could appear in Iran

OpenAI’s Pentagon deal and defense partnerships could place its models in targeting workflows, drone defense systems, and military administration tied to the Iran conflict. The company’s role reflects a broader push to weave generative Artificial Intelligence into US military operations.

Contact Us

Got questions? Use the form to contact us.

Contact Form

Clicking next sends a verification code to your email. After verifying, you can enter your message.