ENFR
8news

Tech • IA • Crypto

TodayMy briefingVideosTop articles 24hArchivesFavoritesMy topics

Cerebras $64B IPO, Microsoft Mdash beats GPT-5.5, AI zero-day

AISaturday, May 16, 2026· 9 videos

Briefing

Audio player
0:00 / 0:00

Cerebras IPO hits $64B valuation

Cerebras Systems surged to a roughly $64 billion market cap after shares doubled from the $150–$160 range to as high as $350 on debut. The blockbuster IPO reflects intense demand for high-speed AI inference infrastructure. Its wafer-scale engine architecture—using an entire silicon wafer as one chip—continues to differentiate it from GPU competitors. Investors appear to be prioritizing raw speed gains, even at significantly higher costs.

Microsoft Mdash beats GPT-5.5 benchmark

Microsoft’s Mdash achieved 88.45% on the CyberGym benchmark, outperforming GPT-5.5 (81.8%) and Anthropic Mythos (83.1%). The system uses over 100 coordinated agents instead of a single model, handling tasks like validation, auditing, and exploit proof. It processed 157 vulnerability tasks across 188 open-source projects, demonstrating strong real-world capability. The results suggest multi-agent orchestration may outperform frontier models alone.

Google flags AI-generated zero-day exploit

Google confirmed a contained attack using a zero-day vulnerability with exploit code generated entirely by AI. The target was an unnamed open-source web application, with mitigation deployed before widespread damage. Investigators believe non-state actors likely used publicly accessible models. The case highlights how AI can dramatically accelerate vulnerability discovery and exploitation timelines.

ChatGPT for Excel audits financial models

ChatGPT for Excel introduces embedded auditing tools to detect errors in spreadsheet-based financial models. The system maps formulas, flags broken references, and categorizes risks between automated fixes and human judgment. It emphasizes traceability, documenting every change made during cleanup. The goal is to reduce decision risk in forecasts and board-level reporting workflows.

Private AI stacks gain enterprise traction

Companies are increasingly deploying local AI systems as internal “second brain” platforms to protect sensitive data. Tools like Open WebUI allow organizations to run ChatGPT-like interfaces connected to private or on-prem models. These setups prioritize data sovereignty and reduce exposure to external training pipelines. The shift reflects growing compliance and confidentiality concerns with cloud AI.

Nvidia opens access to 157 models

Nvidia is offering access to over 157 AI models عبر its ecosystem, including Claude, Mistral, Qwen 2.5, and DeepSeek V3 variants. The program allows experimentation without contributing data for training. While free tiers may introduce latency, they significantly lower barriers to entry. This expands multi-model workflows across enterprises testing diverse providers.

France launches nationwide AI literacy week

France will host “Semaine de l’IA pour tous” from May 18–24, aiming to broaden public understanding of AI. The initiative is backed by the Direction Générale des Entreprises with partners like Mednum and Make.org. Local “Café IA” events encourage informal, community-led discussions. The decentralized model focuses on accessibility and grassroots participation.

AI reshapes work and social dynamics

Workplaces are shifting toward voice-driven AI interaction, with employees dictating prompts to tools like coding assistants. This change is altering office norms, raising concerns about noise, privacy, and collaboration. At the same time, fears are growing about entry-level job displacement and widening inequality. Broader debates are emerging حول who benefits from rapid AI acceleration and who is left behind.

Videos covered

Previous briefings · AI