IT / AI / QA Weekly #1 — Anthropic, Cursor, Meta Glasses and the Real Cost of AI

This week in AI and tech was a good reminder that modern products no longer break only because of bugs. They can break because of regulation, pricing, model access, privacy risks, or because teams trust AI too much.

Anthropic, Fable, Mythos, and model access risk

The biggest AI story of the week was the reported US restriction around Anthropic’s Fable and Mythos models. The important part for QA is not only whether one specific model was blocked or limited. The bigger point is that access to a flagship AI model can become a political and regulatory issue almost overnight.

For product teams, this creates a new kind of dependency risk. What happens if your product relies on a model that becomes unavailable in some countries? What happens if the provider changes access rules, disables a capability, or forces identity verification?

QA take: AI features need fallback scenarios. A good test plan should include model unavailable, model degraded, region blocked, rate limited, and provider policy changed.

SpaceX reportedly buys Cursor for $60B

SpaceX reportedly locked in a $60 billion deal to acquire Cursor, one of the most popular AI coding tools. Whether someone uses Cursor, Claude Code, Codex, or another tool, the direction is clear: AI-assisted development is becoming core infrastructure, not just a nice productivity add-on.

This also changes the QA role. More code will be generated, refactored, and shipped with AI help. That does not remove the need for QA. It increases the need for strong review, regression testing, product understanding, and risk analysis.

QA take: AI-generated code should be treated as high-speed code, not automatically high-quality code. The faster teams generate changes, the more important smoke tests, regression coverage, and clear acceptance criteria become.

Meta smart glasses and face recognition risk

Meta smart glasses are a good example of a product where the main question is not only “does it work?” but also “should it work this way?” If glasses can recognize people in the real world, the product risk moves from UX into privacy, consent, abuse prevention, and regulation.

For QA, this kind of feature requires testing beyond the happy path. What happens with false positives? What data is stored? Can the user understand what is being captured? Can another person opt out? What happens in different regions with different privacy laws?

QA take: Privacy-sensitive features need product QA, security thinking, and clear negative scenarios. Testing only the main flow is not enough.

The UK considers stricter social media rules for children

The UK has been discussing stronger restrictions around children’s access to social media, including age-related proposals. For product teams, this is another reminder that legislation directly affects onboarding, account creation, parental controls, regional behavior, and account recovery.

Age verification sounds simple until you start testing it. Users can enter fake dates, travel between countries, use VPNs, lose access to documents, or get incorrectly blocked.

QA take: Age-based restrictions need careful testing across regions, edge cases, appeals, and analytics. A broken verification flow can block real users or let restricted users through.

Microsoft and cheaper AI models

Another important trend is cost pressure. Big companies are looking at cheaper and more specialized models instead of using the most expensive flagship model for every task. That makes sense. Not every workflow needs the smartest possible model.

For QA and engineering teams, this means AI cost testing becomes real. The question is not only “does the output look good?” but also “how much does this workflow cost at scale?” and “can a cheaper model do the same job well enough?”

QA take: AI workflows should be tested for quality, latency, cost, and model selection. Sometimes the best product decision is not the best model, but the most stable and cost-effective one.

MIT study: AI can help now and weaken judgment later

A MIT Media Lab study suggests that AI assistance can improve people’s ability to detect fake news in the short term, but may weaken their independent judgment over time. This is very relevant for QA.

AI is useful when it helps us compare options, find gaps, generate checklists, or challenge assumptions. It becomes dangerous when people stop thinking and just paste AI answers into test cases, bug reports, or product decisions.

QA take: AI should support critical thinking, not replace it. A QA engineer still needs to verify, question, reproduce, and understand the product.

My QA takeaway of the week

AI tools are no longer just tools. They are product dependencies.

That means QA needs to test them like dependencies: availability, fallback behavior, cost, model quality, privacy, compliance, and user impact. The job is not only to check if a button works. The job is to understand how the product can fail when the model, provider, law, pricing, or user trust changes.