Daily AI brief · May 12, 2026
GPT-5.5 Goes Default as OpenAI Opens Its Ads Manager
GPT-5.5 is now the default ChatGPT model with cross-platform memory, OpenAI opened its self-serve Ads Manager to SMBs, Anthropic launched Claude Security in public beta, Copilot Studio gained GPT-5.5 Reasoning, and the U.S. government quietly expanded pre-release AI model reviews.
1. GPT-5.5 Is Now the Default ChatGPT Model — and It Remembers You *(OpenAI / Releasebot)*
GPT-5.5 Instant has rolled out to all ChatGPT users as the new default, replacing GPT-5.3, and enhanced personalization—drawing from past chats, files, and connected Gmail—is live for Plus and Pro users on web.
The rollout includes GPT-5.5 Pro and GPT-5.5 Thinking variants for eligible paid plans, with stronger multi-step reasoning, tool use, coding, and agentic workflow capabilities.
What it means for SMBs / MSPs: If your team uses ChatGPT Business, the model your workflows were tuned on has already been swapped out — validate any critical prompts and automations this week before assuming behavior is stable.
2. OpenAI Opens Its ChatGPT Ads Manager — Including to SMBs *(Digiday / AdTechRadar)*
OpenAI's self-serve Ads Manager is now in beta in the U.S., open to SMBs, startups, global brands, and agency holdcos — the biggest expansion yet since ChatGPT launched advertising three months ago.
Category coverage is currently limited to household goods, local services, travel and entertainment, and digital products/education.
Meanwhile,
Anthropic doubled down on its opposing stance: "We want Claude to act unambiguously in our users' interests. So we've made a choice: Claude will remain ad-free."
What it means for SMBs / MSPs: You now have a direct line to advertise inside ChatGPT conversations — but also a concrete reason to coach clients on the emerging split between ad-supported AI (ChatGPT, Gemini) and ad-free AI (Claude). The conversation is no longer hypothetical.
3. Anthropic's Claude Security Hits Public Beta — Code Vulnerability Scanning, No API Setup Required *(Anthropic / Cybersecurity News)*
Claude Security is now in public beta for Claude Enterprise customers, offering codebase vulnerability scanning and proposed fixes powered by Opus 4.7, directly on the Claude Platform — no custom agent build or API integration required.
It generates confidence ratings, severity assessments, and reproduction steps for each finding, then proposes recommended patches so teams can move from triage to code changes while keeping humans in the loop.
Access for Claude Team and Max customers is coming soon.
What it means for SMBs / MSPs: If you manage any custom-developed web apps or internal tools for clients, this is a low-friction way to run a defensive scan — it's baked into Claude Enterprise, so no new vendor contracts yet.
4. GPT-5.5 Reasoning Lands in Copilot Studio — Power Automate Gets Self-Healing Desktop Flows *(Microsoft / Copilot Blog)*
GPT-5.5 Thinking is now available in Copilot Studio early release environments as "GPT-5.5 Reasoning," expanding model choice for agents being built on the platform.
Separately,
Power Automate's 2026 wave 1 updates introduce agent-based features for authoring, optimization, and self-healing, applicable to desktop flows and Copilot Studio-driven cloud flow actions.
Copilot Studio's expanded agent usage estimator now includes Dynamics 365 agents, letting you forecast credit consumption across both platforms in one place to avoid unexpected cost surprises.
What it means for SMBs / MSPs: The cost estimator update is practically useful right now — before expanding any client's agent footprint, use it to model Copilot credit burn so the invoice doesn't shock anyone.
5. U.S. Government Expands Pre-Release AI Model Reviews — Anthropic in Legal Gray Zone *(CNBC / CNN Business)*
The Center for AI Standards and Innovation announced agreements with Google DeepMind, Microsoft, and xAI allowing the U.S. government to conduct pre-deployment evaluations of frontier AI models before public release, assessing capabilities and national security implications.
Anthropic's Claude Mythos Preview — which excels at identifying security weaknesses in software — has been limited to a select group of approved organizations as part of a new cybersecurity initiative called Project Glasswing.
The Pentagon declared Anthropic a "supply chain risk," but Anthropic sued the Trump administration in response, and a federal judge in California blocked the government's effort.
What it means for SMBs / MSPs: If you're deploying Claude in a government-adjacent or defense-contractor environment, document your procurement decision and watch this litigation closely — the regulatory status of Anthropic's tools in federal contexts is actively unsettled.
Claude Security (Public Beta)
Anthropic has opened Claude Security to public beta for Claude Enterprise customers, bringing AI-powered vulnerability detection directly into production codebases without the need for custom tooling or API integrations.
- Who it's for: In-house developers at SMBs, MSP engineering teams, or any AI champion managing custom-built apps
- How long a first test takes: 30–45 minutes for an initial scan of a small-to-medium codebase
- First step:
Access Claude Security directly from the Claude.ai sidebar, or navigate to claude.ai/security.
If you're already on Claude Enterprise, an admin just needs to enable it in the admin console — no new tools, no new contracts.
Add one line to your staff AI rules: no client or personal details in ad-supported AI tools. OpenAI's self-serve Ads Manager launch confirmed that ChatGPT conversations inform contextual ad targeting. Sessions can include health information, financial details and personal matters. Tell your team which tool is approved for sensitive work, and give them a short explanation of why.
*For MSPs:* A one-page comparison of ad-supported AI and Claude Enterprise gives professional-services clients a non-technical reason to switch.
Don't present Claude Mythos as a tool your clients can start using for security.
Anthropic has said Mythos is "far ahead" of other models in cybersecurity capabilities, but the company doesn't feel comfortable releasing it publicly yet and is restricting access to a select group of approved organizations.
Pitching it as an available offering — even to technically sophisticated clients — will burn credibility when they try to access it and can't.
AI is splitting into two clearly opposed business models — and your clients will have to pick a lane.
The ChatGPT Ads Manager launch and Anthropic's explicit ad-free pledge mark the moment that the AI assistant market stopped being undifferentiated.
This is becoming the key differentiator for AI brands: those that become ad platforms and those that don't.
For SMBs and their advisors, this is more than a philosophical debate — it's a data governance and compliance question. The tools your teams use to draft contracts, discuss HR matters, and plan finances are now either feeding an ad-targeting system or they're not.
As Anthropic put it: "We are entering a pivotal time for cybersecurity. AI is compressing the timeline between vulnerability discovery and exploitation. We believe the right response is to make sure defenders have access to frontier capabilities."
The same logic applies to data: the gap between governed AI and ungoverned AI is widening fast, and whoever helps clients understand that gap owns the room.