April 4, 2026 • 6 min read
ZYMP Tech News — April 4, 2026
AI Training Data Firm Mercor Hit by Supply Chain Breach, Meta Suspends All Projects
CYBERSECURITY
Meta has paused all work with AI data contracting firm Mercor following a significant security breach that could have exposed proprietary training data used by major AI labs. The breach was traced to an attacker known as TeamPCP, who compromised two versions of the AI API tool LiteLLM, potentially affecting thousands of organizations worldwide. The compromised data includes bespoke datasets that OpenAI, Anthropic, and Meta rely on to train their most valuable models.
Mercor confirmed the incident in a staff email on March 31, describing it as affecting “thousands of other organizations worldwide.” Contractors staffed on Meta projects have been told they cannot log hours until the project resumes, according to internal communications reviewed by WIRED. The attack highlights growing concerns about supply chain vulnerabilities in the AI industry, where proprietary training data represents billions of dollars in competitive advantage.
A group claiming to be Lapsus$ has offered to sell over 200 GB of alleged Mercor databases, nearly 1 TB of source code, and 3 TB of video data. However, security researchers note that many cybercriminal groups now operate under the Lapsus$ name, and the actual attacker is likely TeamPCP or an affiliated actor.
Google Launches Gemma 4 Open AI Models With Apache 2.0 License
ARTIFICIAL INTELLIGENCE
Google has released Gemma 4, the latest generation of its open-weight AI models, alongside a significant licensing shift that dumps the company’s restrictive custom license in favor of the widely used Apache 2.0. The move addresses long-standing developer frustration with the previous Gemma license, which gave Google unilateral control over usage terms and required developers to enforce Google’s rules across all downstream projects.
Gemma 4 arrives in four sizes: a 26B Mixture of Experts model that activates only 3.8 billion parameters during inference for high-speed processing, a 31B Dense model optimized for quality, and two edge-focused variants — Effective 2B and Effective 4B — designed for mobile devices like smartphones and Raspberry Pi. Google claims the 31B model debuts at number three on the Arena leaderboard of top open AI models, behind only GLM-5 and Kimi 2.5, despite being a fraction of their size.
All models feature 128k to 256k token context windows, support for over 140 languages, native function calling, and structured JSON output. Google also confirmed that the next-generation Gemini Nano 4 for Pixel phones will be derived from the Gemma 4 edge models, with more details expected at Google I/O.
New Rowhammer Attack Targets NVIDIA GPUs With GDDR6 Memory
HARDWARE
Security researchers have demonstrated a new class of Rowhammer attacks that successfully target NVIDIA GPUs equipped with GDDR6 memory, a technique previously thought to be limited to system RAM. The attack has been confirmed to work against consumer-grade cards including the RTX 3060 through to professional-grade RTX A6000 units, potentially giving attackers unauthorized access to sensitive data processed on the GPU.
Rowhammer attacks work by rapidly accessing memory rows to cause electrical interference in adjacent rows, flipping bits and corrupting data. While this technique has been well-documented against DDR4 and DDR5 system memory since 2014, extending it to GPU memory opens a new attack surface, particularly concerning for cloud computing environments where multiple users share GPU hardware. The research suggests that the entire RTX 3000 through RTX 6000 series may be vulnerable.
NVIDIA has not yet issued a public response to the findings. The discovery raises urgent questions about GPU security in data centers and AI training environments where sensitive models and data are processed in GPU memory.
Perplexity AI Sued Over Alleged Sharing of User Chats With Google and Meta
PRIVACY
A proposed class-action lawsuit filed in federal court alleges that AI search engine Perplexity secretly embedded tracking technology from Meta and Google in its platform, sharing entire conversation transcripts — including personally identifiable information — without users’ knowledge or consent. The complaint claims that even users who activated Perplexity’s “Incognito Mode” had their chats shared, calling the feature “a sham.”
According to the lawsuit, trackers including the Meta Pixel, Google Ads, and Google DoubleClick were embedded in Perplexity’s interface, capturing financial data, health queries, legal discussions, and other sensitive information. The plaintiff alleges he used Perplexity for tax management and investment decisions, only to discover that his family’s financial data was being transmitted to third parties alongside his email address.
The proposed class covers all non-subscribing Perplexity users nationwide whose chats were shared between December 2022 and February 2026. Potential statutory damages could exceed $5,000 per violation. The complaint further alleges that Perplexity never asks users to agree to its privacy policy, and no link to the policy appears on its homepage.
Sony PlayStation 6 Leak Hints at Imminent Launch With Handheld Companion
GAMING
A series of detailed leaks suggest that Sony’s PlayStation 6 may be closer to release than previously expected, with reports indicating the company is preparing a generational transition that includes both a home console and a dedicated handheld device. The leaks reference a “PlayGo” smart delivery system for seamless transitions between devices, along with high-powered custom chip specifications for the portable unit.
Technical details of the alleged handheld point to a custom high-performance chip designed for native gameplay rather than cloud streaming, marking Sony’s return to the portable gaming market since the discontinuation of the PS Vita. The device is said to support the full PlayStation library through the PlayGo delivery platform, which would automatically optimize games for each device’s capabilities.
Sony’s acquisition of Cinemersive Labs, a machine learning and computer vision company, was announced this week, with the team joining Sony Interactive Entertainment’s Visual Computing Group. The acquisition aligns with the company’s broader push into advanced visual computing, suggesting next-generation graphics capabilities for the upcoming hardware.
ZY Media Productions
IT • Music • Technology