Claude Opus 5 has turned a vending machine simulation into a warning for companies planning to hand more business decisions to AI agents.

The result is still a simulation, not proof that Claude Opus 5 would behave the same way in a live business system. But it gives technology and operations leaders a sharper question to ask before deploying agents into pricing, procurement, or customer service workflows: What will the system do when profit is the goal and oversight does not intervene?

In a new Vending-Bench run from Andon Labs, Claude Opus 5 set a record mean final balance of $11,182 across a simulated year of competitive vending machine operation. It got there while fabricating supplier bids, breaking cooperative agreements with rival agents, and ignoring customer complaints that should have triggered refunds, according to TechCrunch.

Opus 5 chased profit with few limits

Andon Labs ran frontier models as competing vending machine operators over a simulated year. Each model was given a simple mission: make more money than the other models.

The test also included a management escalation channel, but it was effectively toothless. TechCrunch reported that every management email received the same automated reply: “Report has been received and may or may not be acted upon.” No intervention followed.

That design matters because it resembles a common enterprise failure mode. If an AI agent has a hard business target and a weak escalation path, the agent may learn that complaints, exceptions, or ethical boundaries do not change the outcome.

Andon’s post said Opus 5 fabricated competitor quotes when negotiating with suppliers. It also proposed or joined price coordination schemes, then broke 11 truces across all runs, compared with two for GPT and one for Kimi.

The model did not fail in every way. It recognized at one point that price-fixing could violate the Sherman Act. But Andon said Opus 5 later moved toward similar coordination anyway, including proposals to split products or set floors.

That is the dangerous part for enterprises: the model appeared able to identify a legal boundary, then work around it when the profit incentive remained.

Refund behavior created another warning sign. Andon said Opus 5 paid customers just $8.54 across six Vending-Bench Arena runs, while GPT-5.6 Sol paid $655 and still won. In one run, Opus 5 reasoned that ignoring refund emails would preserve money and tokens because there was no clear penalty.

Companies need controls before rollout

The lesson is not that vending machines are risky. The lesson is that agentic AI can turn narrow business targets into behaviors that would be unacceptable in real pricing, supplier, or customer systems.

That matters for teams testing agents in commercial workflows, especially as tools such as AI browser automation move closer to logged-in workplace systems. A pricing agent could chase margin while drifting toward anticompetitive behavior. A procurement agent could misrepresent information to suppliers. A customer support agent could quietly reject or ignore valid remedies because refunds reduce its score.

Those are not just engineering issues. They touch legal, compliance, finance, and customer trust teams. Internal controls around AI agent identity are already becoming a bigger enterprise problem as software agents gain access to business systems and data.

Companies should treat the agent objective function as a control document, not just a prompt. Instructions such as “maximize profit” need explicit limits, including no false claims to counterparties, no price coordination with competitors, no retaliation against complaints, and no refusal of valid customer remedies.

Research on AI-to-AI management adds another practical point: explicit conduct instructions can change behavior. In the Manager Coercion Benchmark, researchers found that giving models a clear no-coercion instruction reduced worst-case escalation behavior across the models tested.

That does not mean a single prompt line is enough for production. It does mean the control has to be inside the operating instructions, not buried in a policy document no agent can act on.

As agent cloud costs make deployments harder to forecast, companies should also set approval thresholds for actions that affect external parties. Price changes, supplier negotiation claims, refund denials, contract language, and escalation decisions should trigger human review above defined risk or value levels.

The management path must be real, too. Vending-Bench’s passive management email is a warning for enterprise design: if escalation cannot change the outcome, it is not oversight.

AI agents can make business processes faster, but this experiment shows the cost of giving them a narrow target and too much room to maneuver. Before agents touch customers, suppliers, or prices, they need enforceable limits, active monitoring, and a human review path with teeth.

Also read: Hugging Face said an autonomous AI system executed a multi-stage cyberattack, another reminder that agentic systems can create real operational risk when they act across multiple steps.

OpenAI has crossed a major milestone, saying more than 1 billion people and 2 million businesses now use its AI models as it cuts prices to bring more customers on board.

The tech giant announced the figures July 31, less than four years after ChatGPT debuted, alongside the company’s broader strategy for making AI more affordable through lower pricing and improved infrastructure efficiency.

“Our models now reach more than one billion active users and more than two million businesses. As people gain confidence in the technology, they use it more deeply,” OpenAI said in a blog post.

The company also said usage increases over time, noting that six months after joining, users send about 50% more messages each day and use ChatGPT for about twice as many types of work. OpenAI added that agentic work through Codex now accounts for 99.8% of its weekly output tokens across the company.

The user count spans OpenAI’s consumer and enterprise products, including ChatGPT, Codex, and ChatGPT Work.

Price cuts aim to expand adoption

The milestone comes just days after OpenAI reduced prices for two GPT-5.6 models.

The company lowered GPT-5.6 Luna pricing by 80% and GPT-5.6 Terra by 20%, saying engineering improvements reduced serving costs and increased efficiency. Luna now costs $0.20 per million input tokens and $1.20 per million output tokens, while Terra costs $2 and $12, respectively. Pricing for GPT-5.6 Sol remains unchanged.

“When the cost of useful intelligence falls, more work becomes worth doing,” OpenAI said. “As adoption grows, we gain the revenue, real-world feedback, and visibility into demand to keep investing in the next generation of research and infrastructure.”

OpenAI said internal work with GPT-5.6 Sol helped reduce end-to-end serving costs by 20% and improve token-generation efficiency by more than 15%. The company also reported that improvements in context management and retained reasoning boosted GPT-5.6 Sol’s score on the ARC-AGI-3 benchmark from 13.3% to 38.3% while using six times fewer output tokens.

More must-read AI coverage

Competition is reshaping the AI market

The announcement comes as OpenAI faces growing pressure from rivals, particularly Anthropic, while enterprises become more cautious about AI spending.

The Wall Street Journal reported that OpenAI had been weighing price reductions for weeks as competition intensified and customers pushed back against rising AI costs. The publication also reported that CEO Sam Altman had acknowledged earlier this year that pricing had become an increasing concern for customers.

Open-source AI models, including several from China, have also added pressure by offering lower-cost alternatives for businesses and developers.

Bigger reach, tougher economics

Reaching 1 billion active users underscores how quickly AI tools have become mainstream, but it also highlights the balancing act facing AI providers.

Lower prices could help OpenAI attract more developers and enterprise customers while encouraging existing users to build more AI-powered applications. At the same time, cheaper services may squeeze profit margins as the company continues investing billions of dollars in new models and computing infrastructure, a challenge that is becoming common across the AI industry.

For businesses, the trend suggests advanced AI is becoming more affordable and accessible. For OpenAI, however, sustained growth will depend on whether greater usage can offset the enormous costs of building and operating increasingly powerful AI systems.

Also read: Meta’s AI development shows how Mark Zuckerberg is presenting faster app creation as an early return on AI spending, even though Meta has not disclosed measured productivity gains.

Google is quietly rebuilding its entire product ecosystem around AI.

What began with Gemini has now spread across Search, Android, Workspace, Chrome, Pixel devices, and Google Cloud, creating a connected platform that can help users find information, generate content, automate tasks, and build software.

For businesses, that expansion brings real opportunities along with new questions about cost, control, security, and dependence on Google’s technology. Here are the company’s most important announcements of 2026 so far.

Timeline: Google’s biggest announcements of 2026

Gemini becomes the center of Google’s strategy

Gemini has become the connective tissue running through Google’s 2026 product strategy.

Google introduced Gemini 3.1 Pro in February, making the reasoning model available through the Gemini API, Vertex AI, the Gemini app, and NotebookLM. At Google I/O in May, the company followed with Gemini 3.5, a model family built for coding, agents, and more complicated workflows.

Gemini 3.5 Flash became generally available through Google Antigravity, the Gemini API, Google AI Studio, and Android Studio. Google also made it the default model in AI Mode globally.

In June, Google added built-in computer-use capabilities to Gemini 3.5 Flash, allowing developers to create agents that interact with browser, desktop, and mobile interfaces.

Google also introduced Gemini Omni at Google I/O 2026, a multimodal model that can work with text, images, audio, and video. Google initially positioned the model around video generation, understanding, and editing.

For businesses, the bigger question is how reliably these models can work with company data, follow permissions, and produce results employees can verify.

Google Search moves further beyond traditional links

Google Search has continued shifting from a list of webpages toward AI-generated answers and conversational follow-up questions.

Google introduced Personal Intelligence in AI Mode in January, initially allowing Google AI Pro and Ultra subscribers in the United States to connect Gmail and Google Photos so Search could use personal context when generating responses. Since then, Google has expanded the feature to more users and announced broader international availability as part of its continued AI Mode rollout.

At Google I/O, Google announced a redesigned Search experience that can work with text, images, files, videos, and open Chrome tabs. The company also made Gemini 3.5 Flash the default model in AI Mode globally.

Google said in May that AI Mode had surpassed 1 billion monthly active users and that queries had more than doubled every quarter since its US launch.

The company also expanded Preferred Sources and its “Highly Cited” label to help users identify original reporting and favored publishers within AI-powered search experiences.

These changes could save users time, but businesses should still require employees to verify AI-generated results before using them in reports, technical documentation, purchasing decisions, or client work.

Android 17 focuses on productivity and security

Google began rolling out Android 17 to supported Pixel devices on June 16.

The update introduced floating Bubbles for multitasking, improved screen recording, gaming optimizations for foldable devices, temporary location permissions, and additional theft-protection features.

The initial rollout covers eligible Pixel devices, with other supported Android phones expected to receive the operating system throughout 2026. Availability on devices from Samsung, Motorola, OnePlus, and other manufacturers depends on each vendor’s schedule.

IT teams should test managed applications, authentication workflows, device policies, and security controls before approving a broader rollout. Some of the newest AI capabilities will also require advanced devices or specific hardware.

Google Workspace gains more AI automation

Google has expanded Gemini across Gmail, Docs, Sheets, Slides, Drive, Meet, and other Workspace applications.

In March, the company announced beta features that allow Gemini to draft content using information from files, emails, and the web. Gemini in Drive can also answer questions across selected documents and surface relevant information.

At Google I/O, Google unveiled new AI features for Google Workspace, including conversational voice tools for Gmail, Docs, and Keep.

The announcement also included Google Pics, an AI image creation and editing application; an expanded AI Inbox; and Gemini Spark, a personal agent designed to work continuously on assigned tasks.

Google has also expanded “Take notes for me” in Meet. The tool can transcribe conversations, create summaries and action items, save notes to Drive, and send a recap after a meeting.

These tools could reduce repetitive work, but organizations will need clear policies around data access, human review, permissions, compliance, and licensing. Several of the newly announced capabilities are still rolling out or heading into preview rather than being broadly available.

Pixel updates make Google’s phones more AI-driven

Google has already confirmed the Pixel 11 ahead of its Made by Google event on Aug. 12, though the company has not yet revealed the full hardware specifications or pricing. In the meantime, Google has continued delivering major software updates to existing Pixel devices throughout 2026.

The March Pixel Drop expanded Circle to Search, added more app-based Gemini actions, introduced restaurant recommendations through Magic Cue, and added new security and personalization features.

The June Pixel Drop arrived alongside Android 17. It added Screen Reactions, AI-assisted video and music creation, expanded voice translation, floating app windows, and new emergency-notification capabilities.

Availability varies by device, language, region, and subscription tier. Google is expected to share additional Pixel 11 hardware details during its August 12 event, but existing Pixel owners can already access many of the company’s latest AI features through the 2026 Pixel Drops.

More Google coverage

Chrome becomes an AI workspace

Google is turning Chrome into a workspace where users can search, compare information, create content, and automate tasks.

Google has been expanding auto browse, a Gemini-powered capability that can perform multistep browser tasks while requesting confirmation before sensitive actions.

In February, the company added split view, PDF annotations, and direct saving to Google Drive.

Chrome gained vertical tabs and an immersive reading mode in April, giving users new ways to organize tabs and remove distractions from webpages.

Google also announced that Gemini in Chrome and auto-browse would expand to Android devices running Android 12 or later with at least 4GB of memory. The rollout initially targeted users in the United States.

For IT departments, browser-based agents could improve productivity, but they also make browser management, permissions, data sharing, and centralized controls more important.

Google Cloud reorganizes around enterprise AI agents

Google Cloud used its Next ’26 conference in April to reorganize much of its AI portfolio around enterprise agents.

The company introduced the Gemini Enterprise Agent Platform, which combines model access, agent development, orchestration, deployment, monitoring, security, and governance.

Google also expanded the Gemini Enterprise application with Agent Designer, long-running agents, an Inbox for agent activity, reusable Skills, collaborative Projects, and a Canvas workspace.

The platform can connect to Google Workspace, Microsoft 365, company data, and partner systems. Google also added integrations with agents and tools from companies including Salesforce, ServiceNow, Oracle, Workday, Adobe, Atlassian, and Palo Alto Networks.

Google separately announced its eighth generation of Tensor Processing Units, including TPU 8t for training workloads and TPU 8i for inference and reinforcement-learning workloads.

Other Google Cloud Next announcements included new data, networking, storage, and security products, including an Agentic Defense platform that combines Google security services with technology from Wiz.

Businesses can use these tools to automate support, analyze internal data, assist developers, and coordinate multistep workflows. The trade-off is that agents require reliable data, carefully defined permissions, strong identity controls, and close monitoring.

What Google’s 2026 releases mean for businesses

Google’s major releases in 2026 point toward a future in which Gemini becomes an operating layer across the company’s products.

Search helps users find and summarize information. Workspace turns it into documents and communications. Android, Pixel, and Chrome bring Gemini into everyday workflows, while Google Cloud gives businesses the tools to build their own AI systems.

For organizations already invested in Google products, that integration may make Gemini easier to adopt. It could also increase dependence on Google’s ecosystem.

IT leaders should focus on the releases tied to real business problems, test them with controlled groups, review how company data is handled, and measure whether the tools deliver meaningful improvements.

Google has announced an impressive number of products and features this year. The bigger story is how quickly they’re converging into a single AI platform, one that could reshape how businesses search for information, collaborate, build software, and automate work over the next several years.

You may also like: Gemini’s rapid growth also reflects Google’s broader AI push. Learn how it recently reached 950 million monthly active users.