2 weeks Ago
Perplexity CEO and cofounder Aravind Srinivas. I n February , days before the Super Bowl, Aravind Srinivas, CEO of AI search startup Perplexity, gave users a sparkly incentive to install its app. He posted on X that instead of buying a Super Bowl ad, the company would give $1 million to one lucky user who downloaded Perplexity’s app, referred it to their friends and asked it five questions during the game, a move to lure more people for a chance to win the grand sum.
“Ask like a millionaire,” he said. But the company’s Android app, which offers not only search capabilities but also acts as an AI assistant, is riddled with a host of security issues that could expose its users to data theft, account takeovers and impersonation attacks from malicious hackers, according to a report by India-based mobile security company Appknox . One of these gaps also lets anyone access Perplexity’s API for free, exposing the company itself to the risk of losing revenue.
Security researcher and Appknox CEO Subho Halder said it’s easy to make clones of Perplexity’s Android app because its code is embedded with what’s called “hardcoded secrets” — sensitive information like passwords and API keys (a string of alphabets and numbers that is used to identify and verify an application making requests to use that API), which can be extracted by an attacker. The cloned app can then be used to trick users into believing it’s the real one, enabling hacks to collect private data like login information and uploaded documents. Perplexity rolled out its agent-like AI assistant for Android devices in January, which it claimed could carry out tasks like booking an Uber, playing a video on YouTube, finding songs on Spotify and making reservations all on its own.
But the slew of security flaws has been uncovered just as Perplexity, reportedly in talks to raise funding at an $18 billion valuation, tries to find new ways to distribute its mobile app to more users and put it in more people’s hands. The company is in talks with smartphone manufacturing giant Samsung to integrate its AI assistant into their phones and it has already reached an agreement with Lenovo-owned Motorola to do the same, according to Bloomberg . Perplexity did not respond to a request for comment.
Perplexity’s app is also susceptible to an attack called “task hijacking” in which a rogue app takes control of the phone’s actions without your knowledge as you use a different one. The now-malicious app can then monitor your activity and collect data. For example, someone could hack Perplexity’s app so that if you’re typing a prompt into Amazon’s search box, it could unknowingly give hackers access to it.
Halder said it could even fall prey to network-based attacks where people on an unsecured network such as an airport hotspot can have their conversations with Perplexity intercepted and their data stolen. Founded in 2022, Perplexity ’s first product was a conversational AI search engine that crawls the web for information and uses a mix of large language models from OpenAI, Anthropic and Meta to answer questions on any given topic by producing AI-generated summaries that include links to sources from across the web. It has raised a total of $900 million in venture funding from tech bigwigs like Amazon founder Jeff Bezos and OpenAI cofounder Andrej Karpathy and is currently valued at $9 billion, according to Pitchdeck.
Perplexity’s app has more than 10 million downloads on Google Play. Security vulnerabilities are just part of the problem for Perplexity. The company has come under fire from Forbes and other media outlets for allegedly plagiarizing their reporting and redistributing it across multiple platforms through a feature called Perplexity Pages.
At the time, Srinivas said that its republishing product feature had “rough edges” and that Perplexity was “improving it with more feedback.” In June 2024, Forbes sent a cease-and-desist letter to Perplexity, accusing it of infringing copyright, to which the Perplexity responded saying the claims were meritless and that factual information is not protected by copyright law. Safety in the world of AI often focuses on the models themselves–ensuring that they’re producing accurate information and aren’t affected by bias.
This report underscores the idea that securing the application where people interact with the models is just as important, Halder told Forbes . Halder’s advice to users is to remove Perplexity’s Android app from the phone until the issues are resolved. AI applications are being built at a breakneck speed and many are failing on the most basic vulnerability checks, Halder said, but “Perplexity is a full-blown security hazard.
” MORE FROM FORBES.
