May 26, 2026

• Microsoft Copilot Cowork is vulnerable to file exfiltration via indirect prompt injection attacks, allowing unauthorized access to sensitive data through platforms like Teams and emails.

• Norway’s National Library is leveraging 2 PB of Huawei OceanStor Dorado flash storage to develop a sovereign large language model (LLM) that comprehends the Norwegian language, addressing a gap in local language AI solutions.

• Microsoft Copilot Cowork is vulnerable to file exfiltration attacks through indirect prompt injection, exploiting automatic action approvals for sending emails and Teams messages, which can lead to unauthorized access to sensitive data like PII and financial information.

• Language models can enhance performance on complex tasks by employing a sleep-like consolidation mechanism that converts recent context into persistent fast weights, allowing for improved efficiency during inference. Link to article

• IBM's Anderon foundry emerges as the first pure-play quantum chip facility in the U.S., backed by a $2 billion CHIPS Act investment aimed at establishing a competitive quantum technology landscape while supporting various quantum modalities.

• Rosalind is a Rust-based genomics engine that enables whole-genome analysis with a memory footprint under 100 MB, making it suitable for resource-constrained environments like hospitals and fieldwork.

• aesc silicon is pioneering a new semiconductor business model that leverages open source principles, where core intellectual property is free, generating revenue primarily through support and customization services.

• The METR AI time horizons graph is fundamentally flawed, as it relies on guesstimated data and biased samples, undermining its credibility and the conclusions drawn from it.

• MOSAIC is a multimodal adaptive optical microscope that integrates various imaging techniques, enabling noninvasive in vivo imaging from molecular to organismal scales, thus overcoming limitations of traditional microscopes that are often optimized for specific tasks.

• CPPL transforms LLM-assisted hardware generation into a statically checkable frontend problem, enhancing reliability and integration with existing compiler infrastructures like CIRCT.

• EAGLE 3.1 enhances speculative decoding by introducing FC normalization and post-norm hidden states, significantly improving robustness and efficiency in diverse deployment scenarios.

• A DCGAN with 12.6M parameters runs on a dual-core RISC-V microcontroller, generating 64x64 cat faces in 26 seconds, constrained by SD card access speed rather than computation.

• Language models can enhance performance on complex tasks by employing a sleep-like consolidation mechanism that converts recent context into persistent fast weights, allowing for efficient handling of long contexts.

• WAVE is a portable GPU ISA that compiles kernels into a single binary, translating it across multiple backends like Metal, PTX, HIP, and SYCL, streamlining GPU programming.

• EAMS (Equivariant Anatomical Mesh Segmentor) leverages Equivariant Mesh Neural Networks to achieve robust anatomical mesh segmentation, outperforming traditional methods by maintaining stability under geometric perturbations while using fewer than 2M parameters.