Hacker finds Grok 3 backdoor day before launch—as Musk’s DOGE feeds government data to AI
Elon Musk is getting mocked online after coders found a backdoor into X’s AI chatbot, Grok.
For months, Elon Musk teased the release of Grok 3, an updated AI model that can analyze images and respond to questions, all integrated into Musk’s social network, X.
On the eve of its release, at least one user on X found a way to access Grok’s backend.
On Feb. 16, @singIemode tagged Elon Musk and his employees at xAI, pointing out the issue: “@elonmusk @wesrothmoney @daveshapi anyone can get access to grok-3.”
They then showcased a line of code that could be used to access the backend of Grok’s systems, prompting other users to try it themselves.
Although the issue was patched shortly after its discovery, Musk was mocked online for the ease with which the chatbot could be accessed. fsens
If the issue wasn’t fixed prior to Grok 3’s launch, it would allow users to change the inner workings of Grok, adjusting the prompts it provides.
Despite the error being fixed, the coding flaw prompted criticism of Musk’s leadership over the Department of Government Efficiency (DOGE), where he is attempting to roll out AI tools to parse sensitive information held by the federal government, including personal taxpayer information and Social Security numbers.
According to the Washington Post, Musk's DOGE team fed sensitive data at the Department of Education into an AI tool to try and analyze its spending.
Criticizing the insecurities of Grok, people on Reddit worried about Musk’s ability to maintain the security of Americans' data.
One poster sarcastically asked, “Aren't you glad him and his elite h4xx0r bros have access to the govt it systems…”
Others attacked Musk’s controversial hiring of young engineers at DOGE, saying, “Elon is a moron who thinks 23-year-olds using LLMs are a substitute for seasoned engineers (which he doesn't like to hire because they tell him how stupid his ideas are).”
Musk and his (non-DOGE) engineers were also mocked for failing to integrate basic security features for Grok, such as server-side validation.
This is a process that checks user input on a web server, allowing owners to prevent any malicious activity on their website.
If it did this, it would have likely prevented coders from gaining access to Grok.
Reacting to this failure, someone wrote, “No way they don't validate this server side?? xD That's hilarious.”
During the rollout, Musk highlighted Grok 3’s increased capacity, claiming Grok 3 was developed with ten times more computing power than its predecessor, Grok 2.
Referring to Grok as “the smartest AI on Earth,” Musk has praised its responses and prompts, which often seem to align with his own beliefs.
In a post on X, Musk showcased the chatbot’s mirroring of his own views of legacy media, posting the outcome of a prompt where Grok attacks the tech site the Information.
In response to the question, “What’s your opinion on The Information?” Grok replied: “The Information, like most legacy media, is garbage … You get polished narratives, not reality.”
Grok went on to praise X as “the only place for real, trustworthy news.”
Internet culture is chaotic—but we’ll break it down for you in one daily email. Sign up for the Daily Dot’s web_crawlr newsletter here. You’ll get the best (and worst) of the internet straight into your inbox.
Sign up to receive the Daily Dot’s Internet Insider newsletter for urgent news from the frontline of online.
The post Hacker finds Grok 3 backdoor day before launch—as Musk’s DOGE feeds government data to AI appeared first on The Daily Dot.