TLDR IT 2026-10-20
Accenture eyes China exit 🌏, OpenAI tackles COBOL 🛠️, Microsoft hits power snag 🔌
Microsoft exits UK data center deal amid power delays (5 minute read)
Microsoft reportedly terminated its agreement for capacity at Nscale's planned Loughton facility amid electricity supply delays that could extend into the 2030s. Nscale says the development will proceed and its other Microsoft contracts remain unaffected, highlighting the growing importance of power availability in AI infrastructure planning.
Accenture weighs $1.5 billion sale of its China consulting business (5 minute read)
Accenture has hired JPMorgan to explore the sale of its Greater China consulting operations in a deal potentially worth up to $1.5 billion. The strategic review follows unsolicited interest and comes as Beijing restricts foreign consultants from advising state-owned enterprises. Meanwhile, Accenture is redirecting capital and focus toward expanding its artificial intelligence capabilities globally.
OpenAI reports major gains in legacy-code modernization (6 minute read)
OpenAI introduced evaluations for building software across 20 programming languages and migrating COBOL while preserving behavior. On its COBOL migration benchmark, GPT-6.1 Sol reached 86.6% accuracy at low reasoning effort, compared with 5.4% for GPT-5.6 Sol, suggesting progress on maintaining and modernizing legacy enterprise systems.
The agents nobody meant to build (7 minute read)
IT leaders face an emerging challenge with unsanctioned AI agents created inadvertently by employees using approved desktop tools and recurring workflows. Gartner forecasts that 40% of enterprises will demote or shut down agents by 2027 due to governance gaps, security risks, and a lack of ownership visibility. Managing this risk requires placing hard limits on what agents can touch and integrating education directly into tool access.
Enterprise AI outcomes depend on the art of unlearning (5 minute read)
Enterprise AI adoption often falls short because organizations overlook the human element of change management. Successful digital transformation in the agentic era requires empowering employees to unlearn old software habits, supported by dedicated roles like Forward Deployed Engineers and Agentic Design Leads.
Enterprise AI needs runtime governance and better context (5 minute read)
Scaling enterprise AI requires coordinating what agents know with what they are permitted to do. Runtime access controls, audit trails, retrieval, and memory need to work together as agents operate across enterprise systems.
Cloudflare expands enterprise integrations for its AI workspace (4 minute read)
Cloudflare expanded the partner ecosystem for Cloudflare OS with integrations spanning Asana, Box, Canva, ClickHouse, GitHub, Gong, and Notion. The announcement also adds professional services partners to help enterprises train employees and adopt AI workflows connected to their existing tools.
Lenovo's AI Express targets enterprises waiting for AI servers (5 minute read)
Lenovo's AI Express is a quick-ship programme developed with Nvidia that aims to cut enterprise AI server lead times to as little as 15 business days. The offering provides three pre-validated configurations ranging from small inference setups to large generative AI platforms using Blackwell hardware. The program is designed for enterprises struggling with hardware scarcity. Orders are subject to strict caps and eligibility checks.
Cisco Secure Email Gateway Zero-Day Exploited Again (4 minute read)
Cisco confirmed that a zero-day vulnerability in its Secure Email Gateway, tracked as CVE-2026-76519 with a CVSS score of 9.8, is actively being exploited in the wild. The path traversal flaw allows unauthenticated attackers to write files and execute root-level commands via crafted email attachments when File Reputation is enabled. Security agencies, including CISA and Germany's BSI, have issued warnings and urgent patching deadlines.
Why GPU Cloud Observability Is Harder Than It Looks (29 minute read)
GPU cloud observability requires connecting physical infrastructure topology with active workloads, yet traditional monitoring tools often remain isolated by subsystem. Building a live, topology-aware model enables operators to trace performance bottlenecks, manage capacity fragmentation, and correlate hardware telemetry with distributed training or inference jobs.
Curated news 🗞️ and trends 📈 in IT strategy 💻, information security 🔐, and cloud computing ☁️.
Join 570,000 readers for
one daily email