Google Cloud Document AI has some worrying security flaws

Cyber security Cloud computing blue abstract digital binary code background. Innovative technology and Artificial intelligence concept. New futuristic system technology symbol. Vector illustration.
(Image credit: Shutterstock / MiniStocker)

Document AI, a Google Cloud service for file processing, had a worrying security flaw which allowed threat actors to steal sensitive data from people’s cloud storage accounts, and possibly even smuggle malware inside, experts have warned.

A report from cybersecurity researchers Vectra AI, who found and reported the flaw to Google in early April 2024, noted how Google Cloud Document AI is a suite of machine learning tools that automates the extraction, analysis, and understanding of documents. It processes unstructured data like invoices, forms, or contracts by converting them into structured, usable information. The service is designed to improve document workflows, enhancing speed and accuracy in data extraction.

Users can process documents stored in Google Cloud via so-called batch processing - automation of document analysis for large volumes of documents simultaneously. During this process, the service uses “service agent”, a Google-managed service that acts as the identity in the process. However, instead of using the caller’s set of permissions for the job, batch processing uses the service agent permissions, which are too broad.

Batch processing woes

As a result, the caller (which could be a malicious individual) can access any Google Cloud Storage buckets within the same project, and through it - all of the data found there. The researchers demonstrated a Proof of Concept to Google, showing how the vulnerability could be abused to exfiltrate a .PDF file, modify it, and then return it to the same place.

Soon after learning about the issue, Google apparently released a patch, and changed the status of the problem as ‘fixed’. However, the researchers said the fix wasn’t sufficient, and pressured the company further. Finally, in early September 2024, Google confirmed applying a downgrade that sorted it out, “because the attacker needs to have an access to an impacted victim's project.”

Via The Register

More from TechRadar Pro

TOPICS

Sead is a seasoned freelance journalist based in Sarajevo, Bosnia and Herzegovina. He writes about IT (cloud, IoT, 5G, VPN) and cybersecurity (ransomware, data breaches, laws and regulations). In his career, spanning more than a decade, he’s written for numerous media outlets, including Al Jazeera Balkans. He’s also held several modules on content writing for Represent Communications.

Read more
A hand reaching out to touch a futuristic rendering of an AI processor.
Google Cloud unveils new AI Protection security tools, no matter which model you use
Shadowed hands on a digital background reaching for a login prompt.
A flaw in Google OAuth system is exposing millions of users via abandoned accounts
Shadowed hands on a digital background reaching for a login prompt.
Private API keys and passwords found in AI training dataset - nearly 12,000 details leaked
An abstract image of digital security.
Identifying the evolving security threats to AI models
Location Data
Cloudflare CDN flaw could expose user location simply by sending an image
Flags of Iran, China, Russia and North Korea on a wall. China North Korea Iran Russia alliance
Cybercrime is helping fund rogue nations across the world - and it's only going to get worse, Google warns
Latest in Security
ransomware avast
Billions of credentials were stolen from businesses around the world in 2024
ID theft
Hackers claim Orange attack, threaten to leak 1TB of data
A computer file surrounded by red laser beams
Free online file converters could infect your PC with malware, FBI warns
Close up of a person touching an email icon.
Criminals are using CSS to get around filters and track email usage
DeepSeek on a mobile phone
More US government departments ban controversial AI model DeepSeek
Ransomware
Fortinet firewall bugs are being targeted by LockBit ransomware hackers
Latest in News
Adobe AI agents
Adobe launches 10 new AI agents to automate key marketing workflows
The Samsung Galaxy S25 Edge on display the January 22, 2025 Galaxy Unpacked event.
Leaked Galaxy S25 Edge pricing gives us a clearer idea of how the super-slim phone will fit into Samsung's lineup
Samsung Galaxy Z Flip 6 in blue
The Samsung Galaxy Z Flip SE may launch months after the Galaxy Z Flip 7
ransomware avast
Billions of credentials were stolen from businesses around the world in 2024
iPhone 12
The iPhone 17 Air could come with a key charging benefit, new leak claims
Nvidia GTC 2025
Nvidia, Google, and Disney's AI-powered Star Wars robot is absolutely the droid I've been looking for