PPalo Alto Networks has unveiled several updates to Prisma AIRS: protection of interactions with Claude Enterprise, analysis of the components used by agents and control of their exchanges with tools.[1] The stakes are concrete for businesses: preventing an assistant from passing on confidential information or following a malicious instruction it encounters while working. These controls should support the use of AI without being mistaken for a guarantee of absolute security.
An integration with Claude already described in August
The 15 October announcement highlights a direct integration with Claude Enterprise covering Claude Code, Claude.ai and Claude Cowork.[1] However, this connection had already been presented in a vendor article dated 5 August 2026.[2] The October news should therefore be read as a set of security updates rather than the first appearance of this integration.
The mechanism relies on “inference hooks”, checkpoints before the model processes content. According to the description published in August, Claude sends the content to Prisma AIRS, which assesses it against the configured security profile and returns a decision to allow or block it.[2]
This architecture places the check before inference, that is, before the model runs on the request. It does not mean that the data stays on the employee’s device: the inspection flow is itself part of the scope to examine during integration.
Protecting data and spotting hijacked instructions
Palo Alto Networks says it reuses its data loss prevention (DLP) policies to control, among other things, personal information, proprietary code and sensitive credentials. The vendor also describes mechanisms for detecting prompt injections and attempts to bypass rules.[2]
A prompt injection seeks to pass off untrusted content as an instruction the assistant should follow. A document to be summarised might, for example, contain a directive telling it to ignore its task. The difficulty lies in distinguishing information that should be analysed from instructions that must never override the work requested.
Three points of control along the agent’s path
This editorial grid distinguishes between the announced functions and the checks useful for assessing them.
| Component | Operational effect | Control question |
|---|---|---|
| Components | Analyse models and skills before deployment.[1] | Which versions and dependencies have been examined? |
| Interactions with Claude | Allow or block content before inference.[2] | Which data triggers a block, and how can this be checked? |
| MCP exchanges | Inspect tool inputs and outputs.[1] | Do all relevant flows actually go through the control? |
The software supply chain now includes skills
The announcement extends AI software supply chain analysis to agent skills and introduces controls designed to identify malicious code before deployment.[1] A skill can bundle instructions, resources and code to help an agent carry out a task. Adding one therefore changes what the system can do and the components it relies on.
Reviews must remain tied to a specific version. Updates and any new access requested must be checked.
Tool calls are another point of control
Prisma AIRS AI Gateway is designed to inspect MCP exchanges in both directions, block certain hidden instructions, mask sensitive data in outputs and enforce access policies.[1] MCP is a protocol that allows AI applications to connect to tools and resources.
For SMEs, testing a development assistant on a limited scope
An SME using Claude Enterprise could start with a pilot project using dummy data. It would test whether a fake secret placed in the context is blocked, then how the system behaves when faced with an unwanted instruction in a document. This scenario is a suggested qualification approach, not a published customer result.
The assessment should measure the threats detected, the legitimate requests blocked and the delay added to exchanges. It should also specify what happens if the inspection service becomes unavailable. The announcements consulted provide neither an independent detection rate nor a detailed availability timetable for each feature.
Professionals fine-tune the rules and check their effect
This development calls directly on Cybersecurity skills. Professionals must identify the information to be protected, define the access required and translate company rules into verifiable controls. A policy that is too vague can let sensitive data through or needlessly hinder work.
Their role includes preparing tests and analysing alerts. A block does not in itself prove that an attack was prevented; nor does the absence of an alert show that an interaction was safe. Results must be reviewed with users and application owners.
New tools and skill updates can change the risks. Controls must be maintained, permissions reviewed and incident handling organised, with the ability to take back control.
Protection will depend on how much real usage is covered
The value of Prisma AIRS lies in bringing several controls closer to how tasks actually unfold. Their effectiveness will depend on the flows inspected, the rules configured and maintenance practices. Allowing an interaction guarantees neither the accuracy of the response nor the appropriateness of the action: business validation remains necessary.
Further reading
Four CCCLX reads on component security and the control of agents in the enterprise.
Azure Linux 4.0: is Microsoft preparing for life after Windows Server?
Open-source security scales up: IBM, Red Hat and Palo Alto join forces
GitHub Copilot bulks up for business: coding agents come under IT control
AI voice agents: Oracle details a LiveKit architecture on OCI
Sources
- 01
Gilmartin, J., Palo Alto Networks (15 October 2026). “Agentic Innovation Meets Enterprise Control”. Overview of Prisma AIRS updates for interactions with Claude, the AI software supply chain and exchanges with tools.
- 02
Thellmann, S. and Hanumanula, S., Palo Alto Networks (5 August 2026). “Prisma AIRS: Unified Data Protection for Claude”. Description of the Prisma AIRS integration with Claude Enterprise and the pre-inference controls.
PDFs, presentations, audio: Adobe wants to turn Acrobat into a content factory
Acrobat no longer just wants to open your documents For more than thirty years, Acrobat has mainly been used to read, create, edit, sign and share PDFs. Adobe now wants to give it a much broader role. On 9 September 2026, the company announced new capabilities based...
Salesforce plugs Claude into its CRM: Claudeforce prepares for life after the interface
What if the CRM soon no longer needed to be opened? For decades, using a CRM meant navigating between customer records, dashboards, menus and forms. With Claudeforce, Salesforce and Anthropic want to change this logic: rather than forcing users to go into Salesforce...
Kimi K3, Mistral, DeepSeek: Oracle turns OCI into an AI model hub
Oracle wants to make OCI a one-stop shop for AI Oracle is stepping up its efforts in the AI cloud battle by significantly expanding the catalogue of models available in OCI Enterprise AI. On 4 September 2026, the group announced in particular the integration of Kimi...
Agentic Data Cloud: Google rethinks data for the age of AI agents
AI agents are forcing Google to rethink data architecture AI agents promise to search for information, make decisions and act directly within company systems, but this autonomy depends on one essential condition: accessing the right data and, above all, understanding...
Google Cloud tackles the hidden bill of AI agents
AI agents raise a new problem: how much does a task really cost? As companies move from chatbots to AI agents able to work through multiple steps, the question of cost becomes much more complex. A chatbot generally handles a request and generates an answer, whereas an...
Workday, Navan, Avalara: ChatGPT Work moves into business software
ChatGPT Work is gradually leaving the chat window With GPT-6 Astra, OpenAI is no longer simply trying to improve the quality of ChatGPT’s answers. The company wants to bring its AI closer to the applications in which employees actually work. With the arrival of GPT-6...
From chip to cloud: Groq raises 350 million to scale up
Groq now wants to sell cloud rather than just chips Groq is pursuing a major strategic transformation. Long seen as one of Nvidia’s challengers thanks to its LPUs (Language Processing Units) specialised in artificial intelligence model inference, the group is now...
Muse Code: Meta launches its AI agent to tackle giant codebases
Meta wants to bring AI agents into complex software projects Meta is stepping up its efforts in AI-assisted software development with Muse Code, a new agent designed to work on large codebases, where traditional assistants can quickly lose the context needed to...
Qwen3.8-Max reportedly coded for 16 days without human intervention
Alibaba tests how far an autonomous AI developer can go Alibaba claims to have reached a new milestone in agentic software development with Qwen3.8-Max, its new flagship model launched in early August 2026. In an internal test, the model reportedly worked fully...
Two months, $1.1 billion: River AI is already exciting investors
A spectacular raise for a start-up barely out of the shadows Two months after emerging from stealth mode, River AI has just completed one of the most spectacular funding rounds of the year in artificial intelligence. The start-up founded by Igor Babuschkin, a former...
