Weekly Tech News: Microsoft AI Flaw, Uber Robotaxis, and Anthropic Export Controls
AutoJack Vulnerability: Microsoft Discloses RCE Flaw in AI Agents
Microsoft researchers have uncovered a critical cybersecurity exploit chain dubbed AutoJack, which severely impacts web-enabled artificial intelligence agents. This AI vulnerability allows a single malicious webpage to hijack an AI browsing agent, executing arbitrary code on the host machine without requiring any user interaction beyond a URL click.
By exploiting a combination of missing authentication on Model Context Protocol (MCP) endpoints, unsafe parameter handling, and an origin allowlist bypass, attackers can trigger Remote Code Execution (RCE). The AutoJack flaw specifically targets pre-release builds of AutoGen Studio, effectively turning the AI agent into a malware delivery vehicle by crossing the localhost trust boundary. Maintainers have since hardened the upstream main branch to mitigate this critical AI security risk.
Sources:
Uber Expands Robotaxi Network with Stellantis, Wayve, Nuro, and Lucid
Uber is aggressively scaling its autonomous vehicle network through two major strategic partnerships announced this week. The ridesharing giant has teamed up with Stellantis and Wayve to develop and deploy Level 4 driverless robotaxis globally. This collaboration will utilize Stellantis’s bespoke L4-Ready Platforms alongside Wayve’s cutting-edge AI driving technology.
In a separate autonomous driving initiative, Uber is collaborating with Nuro and Lucid Group to launch a premium robotaxi service in Houston by mid-2027. This Texas-based autonomous fleet will feature 100 Lucid Gravity electric vehicles (EVs) equipped with Nuro’s Level 4 universal autonomy platform. Uber will own and operate the Houston fleet, supported by a newly secured 50,000-square-foot depot facility.
Anthropic Fable 5 and Mythos AI Models Halted Over US Export Controls
Anthropic has officially deactivated public access to its newly launched Fable 5 and Mythos generative AI models following a strict US government directive. The Trump administration classified these advanced frontier AI models as dangerous munitions, utilizing federal export-control authorities to prohibit access by foreign nationals.
Because the AI startup could not reliably differentiate between domestic and international users, Anthropic was forced to take the generative AI systems offline entirely. This unprecedented regulatory action underscores the growing national security concerns surrounding the capabilities of frontier artificial intelligence platforms and global AI regulation.
Sources:
EU Commission Selects EUROPA Consortium for 400B-Parameter Open-Source AI Model
The EU Commission has officially selected the Italian-led EUROPA Consortium to develop a massive open-source frontier AI model. This upcoming generative AI system will feature an impressive 400 billion parameters and is specifically designed to natively support all 24 official languages of the European Union.
This strategic initiative represents a significant push by European regulators to establish a sovereign artificial intelligence infrastructure. By funding this massive open-source AI project, the EU aims to compete directly with proprietary AI platforms developed by major tech corporations.
Sources:
Kubernetes Infrastructure Upgrades: OCI Adds RDMA Support and Spectro Cloud Integrates AI Tools
Recent advancements in Kubernetes infrastructure are heavily targeting artificial intelligence and high-performance computing (HPC) workloads. Oracle Cloud Infrastructure (OCI) has announced that its Kubernetes Engine now supports RDMA-connected Compute Clusters for managed node pools. This critical cloud upgrade enables the low-latency networking required for distributed AI training and multi-node inferencing.
In parallel, Saturn Cloud has partnered with Spectro Cloud to integrate its managed AI platform directly into Palette-managed Kubernetes environments. This powerful integration allows organizations to deploy production-ready AI tools—such as Jupyter notebooks and distributed training capabilities—across their existing Kubernetes clusters, scaling seamlessly from the data center to the edge.
Sources:
Quantum Error Correction: Quantum X Labs Launches Integrated Computing Program
Quantum X Labs Inc. has launched a groundbreaking integrated computing program that combines its proprietary CliniQuantum algorithm for clinical trial data analysis with advanced Quantum Error Correction (QECC) Decoder technology.
This quantum computing initiative aims to maximize computational accuracy and reliability by executing the algorithm across multiple quantum environments, including leading quantum processors and hybrid GPU-quantum systems. By integrating the QECC Decoder, Quantum X Labs seeks to enhance error mitigation and optimize the path toward scalable, fault-tolerant quantum applications. The program is specifically designed to evaluate algorithmic performance and decoder capabilities across diverse computational platforms in the healthcare sector.
Sources: