AI’s next test: useful tools, visible costs, and safer deployment

A MarTech roundup revisits the risks of folding a major collaboration platform into a larger company’s product ecosystem. For builders, it’s a reminder that integration is about preserving the workflows and trust people rely on—not just connecting features.

Anthropic has introduced tools aimed at showing companies what their spending on its models is going toward. Better visibility could help teams connect usage to value—and make it easier to spot costs that aren’t delivering.

Anthropic’s opt-in OSS Scanner applies AI to finding vulnerabilities in open-source software. If tools like this fit naturally into maintainers’ workflows, they could help broaden security review; teams will still need to assess findings and decide what to fix.

DeepSeek is releasing software it uses to operate AI models at scale, with the potential to reduce operating costs for others. For teams building on open infrastructure, this points to efficiency gains that may come from better systems tooling—not just smaller models.

Belgian firm Aikido has released an open-weight AI model for cybersecurity, as interest in locally developed tools grows. Open weights may give organizations more room to adapt deployment to their requirements, though each team will need to evaluate how well it performs for its use cases.

The department has opened three major AI products for employees to use at work. For public-sector teams, the practical question is how access is paired with clear rules for appropriate use and handling sensitive work.

MarketingProfs offers a weekly look back at selected AI news and developments from late June and early July. Roundups can be a useful way to compare signals across the field, especially for people who can’t follow every release as it happens.

The Guardian reports that OpenAI announced a product called “dots” after dropping plans to launch a new model over safety concerns. It’s a reminder that product direction and safety decisions can move on different tracks—and that pauses can shape what users see next.

Google has released three new Gemini models, including Gemini 3.6 Flash and Gemini 3.5 Flash, according to the New York Times. More options give developers additional choices to test against their needs—but the right fit depends on the task, not the model label alone.

Nvidia, SpaceX, Microsoft, and other technology companies have launched an initiative focused on AI safety for open models. For teams sharing or adapting models, the effort highlights the importance of treating safety as part of the surrounding ecosystem, not just the original release.