Artificial intelligence is forcing companies to rethink security faster than many organizations are prepared for. A new TechCrunch report highlights how even Google, one of the world’s most powerful cloud and AI companies, is still navigating the same transition it tells enterprise customers to manage carefully.

The core message is simple: AI security cannot be treated as an add-on. As businesses adopt AI tools, models, agents, prompts, and data pipelines, the attack surface becomes wider, faster, and more difficult to control. That shift is making AI security a leadership issue, not just a technical concern for internal security teams.

Google Cloud’s Warning: Security Cannot Come Later

Francis de Souza, chief operating officer of Google Cloud, told TechCrunch that companies need to take a platform-level approach to AI security. His argument is that organizations cannot leave AI safety, governance, and auditability to individual employees.

One of the biggest risks he highlighted is “shadow AI.” This happens when employees use consumer AI tools without approval, oversight, or proper data controls. For companies, that creates obvious problems. Sensitive data can move into systems the business does not monitor, audit, or govern.

De Souza said AI strategy, data strategy, and security strategy need to move together. In practical terms, that means companies should not rush into AI deployment without knowing where their data lives, who can access it, how models are being used, and what controls exist around those systems.

The Multicloud Problem Makes AI Security Harder

The report also points to a reality many companies underestimate: most large organizations are not truly operating in a single-cloud environment.

Even if a company chooses one main cloud provider, it may still rely on SaaS apps, vendors, business partners, third-party integrations, and external platforms that run across different cloud systems. That makes security more complicated because the company’s data and workflows are spread across multiple environments.

De Souza argued that companies need a consistent security posture across clouds and models. This is especially important as AI systems begin to connect with internal tools, enterprise data stores, customer records, software workflows, and business applications.

The bigger issue is not just where data is stored. It is how AI agents may move across systems and uncover information that was previously forgotten, poorly labeled, or protected by outdated access rules.

AI Agents Could Expose Forgotten Data

One of the most important risks in the TechCrunch report involves AI agents inside enterprise systems.

De Souza warned that many organizations have old data repositories, such as forgotten SharePoint servers or outdated internal systems, where access controls may not have been updated for years. In the past, that may not have caused immediate problems because few people knew those systems existed.

AI agents change that. If agents are allowed to search across an enterprise, automate tasks, or retrieve information from internal systems, they may find sensitive data that employees forgot about. Once surfaced, that data can become exposed to people or systems that should never have had access to it.

This turns legacy data management into an AI security issue. Companies do not only need to protect new AI tools. They also need to clean up old permissions, outdated repositories, stale credentials, and forgotten storage locations.

Attack Speed Has Changed the Security Equation

The TechCrunch article also highlights how quickly modern attacks can move.

De Souza said the average time between an initial breach and the next stage of an attack has dropped sharply, from eight hours to 22 seconds. That kind of speed changes how companies need to think about defense.

Traditional security models often depend on human review, alerts, escalation, and manual response. But if attacks move in seconds, human-led processes may be too slow on their own.

That is why Google Cloud is promoting the idea of AI-native defense. In this model, AI agents help run security operations at machine speed, while humans oversee the process rather than handling every action directly.

AI-Native Defense Is Becoming Part of the Conversation

The idea of agentic security is becoming more relevant as AI systems increase both the risk and the defensive possibilities.

In theory, AI-driven defense can monitor threats, detect unusual behavior, respond faster, and reduce pressure on human security teams. That matters because companies already face a shortage of experienced cybersecurity professionals.

But the shift also raises a difficult question: if AI systems are used to defend companies, who watches the defensive AI? Oversight, governance, logging, auditability, and escalation rules become just as important as the technology itself.

The TechCrunch report makes clear that this is no longer only a security department issue. AI security decisions now affect boards, executive teams, engineering leaders, legal departments, compliance teams, and product managers.

Google’s Own API Key Problems Show the Gap

The report becomes more complicated when it turns from Google’s advice to Google’s own recent security and billing issues.

TechCrunch cites reports from The Register about Google Cloud developers who received large bills after unauthorized API calls to Gemini models. In some cases, API keys originally used for Google Maps became capable of accessing Gemini after Google expanded their scope.

That created serious consequences for developers. TechCrunch notes one case where Rod Danan, CEO of interview-prep platform Prentus, faced a bill of $10,138 in roughly 30 minutes after attackers exploited a compromised API key. Another developer, Isuru Fonseka in Sydney, reportedly faced about AUD $17,000 in charges.

Google refunded both after The Register reported on the cases. Still, the episode shows how fast AI-related abuse can become expensive when credentials, billing controls, and product access change in ways users may not fully understand.

Key Revocation Is Another Security Concern

The TechCrunch report also discusses research from security firm Aikido about what happens when a developer deletes a compromised Google API key.

According to the report, Aikido found that deleted keys may continue working for up to 23 minutes while revocation spreads across Google’s infrastructure. During that window, attackers may still be able to use the key.

The issue matters because many developers assume deleting a compromised key immediately shuts down the threat. If revocation is delayed, attackers can continue making API calls during the gap, potentially increasing costs or accessing data.

The report also notes that newer Google credential formats appear to revoke faster, suggesting the slower process for some older keys may be solvable.

Why This Matters for Companies Using AI

The broader lesson is that AI security is still being figured out in real time. Companies are being told to adopt strong governance, clean data strategies, platform-level controls, and AI-native defense. Those recommendations are serious and practical.

At the same time, even major platform providers are facing their own challenges around API access, billing limits, key revocation, product scope changes, and user communication.

That gap is important. Businesses adopting AI should not assume that platform scale automatically solves security risk. They still need internal controls, permission reviews, key management, spending alerts, vendor scrutiny, and clear rules for employee AI use.

Final Takeaway

The TechCrunch report shows that AI security is no longer a future concern. It is already affecting cloud platforms, developers, enterprises, and security teams.

Google Cloud’s message is that AI requires security from the beginning, not after deployment. That advice is sound. But the recent issues around Gemini API access and key revocation show that even the companies building AI infrastructure are adapting as problems emerge.

The real takeaway is that AI security is now a moving target. Companies cannot rely on old defenses, informal employee behavior, or blind trust in vendors. They need stronger governance, faster response systems, better data hygiene, and clearer accountability before AI becomes deeply embedded in everyday business operations.