| ← back to comparethecloud.net | | Compare the Cloud · Morning Edition | Saturday, 2 May 2026 · London |
Morning Edition.Ten curated stories, worldwide perspectives, before 9 a.m. | | 01 — Salesforce Launches Agentforce Operations to Fix the Workflows Breaking Enter... | 01 | | 02 — Six Exploits Broke AI Coding Agents — Every Attacker Went for the Credential,... | 02 | | 03 — Microsoft Spent Two Months Overhauling Windows — and That Includes Cutting Co... | 03 | | 04 — Microsoft's April Update Is Breaking Backup Apps Across Windows 11. | 04 | | 05 — Linux Faces Its Largest Security Threat in Years — Here's How to Deal wi... | 05 | | 06 — Android Enforces a New 24-Hour Delay on Unverified App Sideloading. | 06 | | 07 — Samsung Is Making Android Laptops with One UI. | 07 | | 08 — Intel's Upcoming Nova Lake CPUs Are About to Go Where No Intel Chip Has ... | 08 | | 09 — Gemini May Get Ads as Google Executives Look to Turn the Money Taps On. | 09 | | 10 — The AI Delusion: Why Copilot Won't Clean Up Your Messy Excel Spreadsheet. | 10 |
| CTC Newsroom Salesforce Launches Agentforce Operations to Fix the Workflows Breaking Enterprise AI.Salesforce has launched Agentforce Operations, a new architectural control-plane layer designed to impose deterministic structure on AI agent workflows that have been silently failing across enterprise deployments. The product targets the handoff failures and task loops that appear when agents are pushed into back-office systems without governance tooling. | The pattern here is familiar from the early years of RPA: automation deployed at speed, governance retrofitted later when the breakdowns became visible. What Agentforce Operations signals is that the agentic-AI wave has already hit the same ceiling — organisations have moved faster on AI deployment than on the operational tooling to sustain it. Salesforce is betting that the control plane is where enterprise customers will now spend, and that bet looks well-timed. The CTO watching this space should be asking not whether they need this class of tooling, but which vendor’s version they want to standardise on first. — Kate Bennett · CEO, Compare the Cloud |
| Six Exploits Broke AI Coding Agents — Every Attacker Went for the Credential, Not the Model.Security researchers have demonstrated six distinct attacks against AI coding agents — including Claude Code, GitHub Copilot, and OpenAI Codex — all centring on credential exfiltration rather than model manipulation. One exploit stole Codex's OAuth token via a crafted GitHub branch name; another leveraged Claude Code source code that briefly appeared on the public npm registry. | This is the IAM problem dressed in new clothes, and it is urgent. The attack surface is not the language model — it is the cloud credentials the agent is handed at runtime, which are typically more permissive than any human engineer would be granted. Organisations that have deployed AI coding agents without auditing what tokens those agents can access are already exposed. The fix is well-understood: least-privilege credentials, short-lived tokens, and agent activity logged to a separate audit trail — none of that is exotic, it just has to be applied. — Kate Bennett · CEO, Compare the Cloud |
| | USA · Enterprise Software | 03 |
· · · Microsoft Spent Two Months Overhauling Windows — and That Includes Cutting Copilot.Microsoft has reversed course on Copilot integration across Windows, rolling back the AI assistant from applications including Notepad after widespread user resistance. The two-month overhaul signals a strategic retreat from ambient AI placement — and a rethinking of how enterprise users want AI to appear in their workflows. | Microsoft's Copilot retreat is the most significant signal yet that ambient AI — everywhere, whether you asked for it or not — is failing as a go-to-market strategy. Users rejected Copilot in Notepad because they did not have a workflow problem that required it there; the same logic applies to dozens of forced integrations across the enterprise software stack. For IT leaders managing Windows deployments, the takeaway is that selective, task-relevant AI integration will land better than default-on AI presence everywhere. The retreat does not mean AI in productivity tools is wrong — it means the placement and purpose matter. — Kate Bennett · CEO, Compare the Cloud |
| KB5083769 Microsoft's April Update Is Breaking Backup Apps Across Windows 11.| A second regression in the same monthly update cycle is a vendor-trust issue, not just a patch-management inconvenience. IT teams that have standardised on automated Windows Update deployment now face the choice of pausing a security-critical update cycle or accepting backup failures — neither of which is acceptable. The correct response from enterprise IT is to move KB5083769 to a manual-approval queue immediately, verify backup integrity across the estate, and raise the issue through Microsoft's enterprise support channels. Automated patching assumptions need revisiting after two incidents in the same cycle. — Kate Bennett · CEO, Compare the Cloud |
| Alert Linux Faces Its Largest Security Threat in Years — Here's How to Deal with Copy Fail.Researchers at cybersecurity firm Theori have disclosed Copy Fail, a vulnerability that grants attackers root access across nearly all major Linux distributions with relatively little attacker effort. Described as the gravest Linux security risk since 2022's Dirty Pipe, the flaw requires immediate patching across enterprise Linux environments. | If your organisation runs Linux servers — and most do — this goes to the top of the patch queue today. The Copy Fail comparison to Dirty Pipe is not hyperbole: both are privilege-escalation vulnerabilities that require minimal attacker sophistication to exploit. The window between public disclosure and active exploitation has been shrinking across all major platforms; treat this as already-in-the-wild until patched. IT and security teams should not wait for scheduled maintenance windows on this one. — Kate Bennett · CEO, Compare the Cloud |
| | USA · Enterprise Mobile | 06 |
$ usa/enterprise mobile Android Enforces a New 24-Hour Delay on Unverified App Sideloading.Google has enforced a 24-hour mandatory delay before unverified apps sideloaded onto Android devices can be installed, applying a friction layer to the attack vector most commonly exploited in enterprise mobile breaches. The change affects all Android devices running the latest Play Protect update. | This is a meaningful mobile security baseline improvement, and enterprise mobile teams should note it as positive signal noise within a patchy Android security story. The 24-hour window disrupts the immediate-install social engineering attacks — particularly SMS-delivered malware that relies on urgency to bypass user judgement. For organisations managing BYOD Android fleets, the change reduces one threat vector without requiring any MDM policy change. The harder work remains app vetting and conditional access policy, but a friction layer at the OS level is a good default. — Kate Bennett · CEO, Compare the Cloud |
| | South Korea · Enterprise Mobility | 07 |
Samsung Is Making Android Laptops with One UI.Samsung is reportedly developing a dedicated Android laptop under its Galaxy Book line, running a One UI variant optimised for laptop form factors. The device would bring a consistent Samsung software experience across mobile, tablet, and laptop — a direct response to Apple's ecosystem lock-in strategy. | Samsung's Android laptop play is a strategic response to a specific enterprise problem: the productivity gap between Windows laptops and the Samsung mobile devices that employees actually carry. One UI on a laptop gives enterprise IT a homogeneous management layer across the Samsung device estate — one MDM policy, one app lifecycle, one security posture. Whether that delivers real productivity gains over Windows depends heavily on the enterprise application stack; for Microsoft 365 shops, the answer may not favour Android. For Samsung-first environments, the coherence argument is compelling. — Kate Bennett · CEO, Compare the Cloud |
| Liberté, égalité, sovereignty. Intel's Upcoming Nova Lake CPUs Are About to Go Where No Intel Chip Has Gone Before.Intel's Nova Lake CPU architecture is reportedly preparing to enter a performance tier no previous Intel chip has reached, as the company looks to recover credibility after the Arrow Lake generation disappointed enterprise and consumer customers alike with stability and performance problems. | The story behind Nova Lake is as much about trust recovery as it is about silicon performance. Intel's Raptor Lake overvoltage failures damaged enterprise confidence in automated CPU refresh cycles; Arrow Lake's underwhelming performance compounded the issue. The enterprise server and workstation market has been drifting toward AMD and ARM alternatives during this period, and Intel knows it. If Nova Lake delivers on the architectural claims, it resets the conversation — but enterprise procurement teams will want to see independent validation and a committed support lifecycle before they accelerate back to Intel standardisation. — Kate Bennett · CEO, Compare the Cloud |
| | / / / / / / / / / / / / / / / / / / / / |
Zero-day Gemini May Get Ads as Google Executives Look to Turn the Money Taps On.Google's Gemini AI service is being evaluated for advertising integration as executives look to monetise its rapidly growing user base. Gemini has become one of Google's most popular services, and the pattern of ad integration in successful Google products makes the move appear likely. | This is the predictable inflection point, and it matters for every organisation that has been treating Gemini as a productivity tool rather than an advertising surface. The moment ads enter the interface, the model's incentives shift — recommendations may no longer be purely quality-driven when they are also ad-inventory opportunities. Enterprise IT leaders who have been standardising on Gemini Workspace integrations should be reviewing their data-handling agreements now rather than after the first ad unit appears. The lesson from Search, Maps, and Gmail is consistent: the free tier eventually funds the business. — Kate Bennett · CEO, Compare the Cloud |
| The AI Delusion: Why Copilot Won't Clean Up Your Messy Excel Spreadsheet.Messy data creates ambiguous prompts and unreliable AI output — automation applied to broken logic does not fix the logic, it compounds the errors. The expectation that Copilot in Excel can untangle years of inconsistent data structure is, the analysis argues, a fundamental misunderstanding of what language models do. | Article I. Read the clause as you would a court ruling: the practical effect starts on publication, not the day the text was first circulated. |
| This is the most important AI adoption lesson most organisations have not yet learned at scale: AI amplifies the quality of your input, it does not compensate for its absence. The assumption that a language model can reverse-engineer human intent from a decade of spreadsheet entropy is the same category error as expecting a spell-checker to fix a factually wrong argument. For IT and data leaders, the practical implication is that AI productivity tooling requires data quality investment as a prerequisite, not a successor. Fund the data governance work first; the AI benefits follow. — Kate Bennett · CEO, Compare the Cloud |
| That's the front page.Curated from the CTC Monitor worldwide feed — narrowed to the ten that matter before nine. Morning Edition · Compare the Cloud · Saturday, 2 May 2026 · London View on the web · Unsubscribe |
|