BIP NYC

collapse
Home / Daily News Analysis / GTT bets that the network, not the log file, is where AI-era security gets won

GTT bets that the network, not the log file, is where AI-era security gets won

Oct 22, 2026  Twila Rosenbaum  77 views
GTT bets that the network, not the log file, is where AI-era security gets won

Security operations have a speed problem, and AI is making it worse. Attackers use AI to find weaknesses and write exploits faster than many human-staffed security teams can respond. Defensive tools often collect telemetry, ship it to a remote platform, normalize it, and then analyze it. Every step adds delay. In security, delay is exposure.

GTT Communications is betting that the answer is to stop moving so much data and put intelligence where network traffic already flows. The company launched GTT Defense Halo, an AI-native network defense platform that runs on an AI factory embedded in GTT's global Tier 1 backbone. The AI factory connects to New York, Dallas, London, and Prague. Each customer receives a dedicated, single-tenant instance rather than a slice of a shared cloud environment.

Key facts at a glance

  • GTT Defense Halo is an AI-native network defense platform running on GTT's AI factory and Tier 1 backbone.
  • The service is delivered as a dedicated single-tenant instance for each customer, with processing nodes in New York, Dallas, London, and Prague.
  • The platform has three parts: Recon for configuration and exposure analysis, Detect for behavioral threat analysis, and Response for approved actions through agentic runbooks.
  • The most valuable data comes from correlating identity, host, user, syslog, and network traffic in real time, not from packet capture alone.
  • Early use cases focus on cross-vendor firewall policy conversion and CVE correlation across large device estates.
  • GTT argues for human decision-making with AI execution, rather than fully autonomous response.
  • The platform targets mid-to-large distributed enterprises, starting with existing managed firewall and network customers.

Why the network is the control point

Most security analytics still follow a hub-and-spoke model. Data is generated at the edge, sent to a centralized analytics stack, and then returned as a finding or recommendation. That model made sense when security teams needed a single repository for logs and when network bandwidth was plentiful relative to telemetry volumes. It makes less sense when AI-driven attacks compress decision windows and when telemetry volumes are growing faster than many budgets.

By placing the AI factory on its own network, GTT says it can ingest and analyze data on-net rather than over the top. That removes egress charges and connectivity limits that can constrain how much telemetry a customer sends to an external analytics platform. It also changes the latency profile. The analysis can occur closer to the traffic being examined, which matters when the goal is to detect and respond at machine speed.

The single-tenant design is another differentiator. A shared cloud service may offer isolation through logical partitions, but a dedicated instance gives each customer its own baseline, its own data boundary, and a clearer story for data residency. That matters in regulated industries where proving where data is processed can be as important as proving what the analysis found.

The architecture also reflects GTT's history. A network operator that owns transport and AI infrastructure can remove some costs and latency that software-only security vendors must pass through or work around. That does not guarantee better security outcomes, but it changes the economics of how much data can be analyzed and how quickly.

Three modules, one operating loop

GTT Defense Halo is organized into three components. Recon provides ongoing analysis of firewall and device configurations, comparing them against known security frameworks to identify compliance violations and security exposures. It also maps known CVEs to the customer's monitored assets and quantifies exposure risk, so security teams can see which vulnerabilities matter in the context of actual traffic and device role.

Detect establishes a baseline of normal behavior for a specific customer's network and performs continuous threat analysis. The goal is to find anomalies that go beyond known CVEs and indicators of compromise. That is important because many breaches do not announce themselves with a known signature. They appear as unusual identity behavior, unexpected host communication, or a deviation from a learned pattern.

Response feeds findings through an agentic runbook into an existing security service and handles approved responses across managed environments. The emphasis is on approved responses. The platform can carry out actions, but it is not designed to make open-ended decisions without human oversight. That distinction is central to how GTT is positioning the product.

The data correlation is the most interesting part of the design. GTT initially expected packet capture and software-defined networking data to be the most valuable inputs. According to the company, they were not. The real value came from correlating identity, host, user, and syslog data with network traffic, all at once and in real time. That conclusion is directionally correct. Breaches rarely appear in a single data source. They appear in the gaps between sources, where siloed tools are weakest.

Real-time correlation and the galaxy model

The demo highlighted a real-time galaxy model of every host-to-host conversation. An operator can click a node to see everything it is communicating with, then move through time to trace where an event started and how it spread. Threat hunting that can take weeks becomes a matter of minutes. The model is not just a visualization. It is a way to make correlated data usable under time pressure.

That matters because security teams are often rich in data and poor in context. A firewall log can show a blocked connection. An identity system can show a successful login. A host sensor can show a process change. None of those facts alone may be decisive. Together, in the right sequence, they can reveal a compromise in progress. The value of on-net analysis is that the correlation can happen while the event is still unfolding.

The platform also uses Nvidia technology. GTT says the partnership is not just branding. The value comes from combining Nvidia GPUs, Morpheus, an AI application framework, and Nvidia's NIM inference microservice model, which GTT extended with its own patent-pending work. The details of that extension are not public, but the combination points to a design that uses accelerated computing for real-time inference rather than batch analysis.

Firewall sprawl is the near-term win

The use case that matters most to early customers is a familiar struggle: real-time analysis of firewall configurations against known security frameworks and building a single policy across multiple vendors. Many large enterprises have accumulated firewalls from different vendors over years of acquisitions, regional requirements, and technology refreshes. Policy drift is common. A rule that made sense in one environment can become an exposure in another.

GTT described an enterprise customer with 110 sites moving from cloud-managed Palo Alto Networks firewalls to on-premises Fortinet gear. Done manually, the conversion took about three and a half weeks. GTT Defense Halo completed it in roughly three and a half hours, with human review. Once the changes were approved, a rerun took 11 minutes. Those numbers are striking, but the important detail is the human review. The platform accelerated the work, but people still approved the changes.

CVE correlation is another practical use case. With AI-assisted bug hunting driving CVE volumes sharply higher, knowing which vulnerability matters on which device, in the context of actual traffic, is far more valuable than another prioritized list. GTT analyzed 145,000 devices across its managed firewall estate, spanning Cisco, Check Point, Palo Alto, HP, and Fortinet, in two hours and 15 minutes. That kind of cross-vendor analysis is difficult to do manually at scale.

GTT has also deployed Defense Halo across its own enterprise estate. The company says it reduced licensing costs and saved more than $1 million after accounting for integration work and manual correlation labor. GTT also says that, for its own use case, Defense Halo could now handle about 98% of what its SIEM had previously done. That is a bold claim, and it is limited to GTT's environment. It is a sign of direction, not a reason for every customer to remove a SIEM next quarter.

The case for less automation, not more

One of the more interesting aspects of the launch is that GTT is arguing for less automation at a time when much of the industry is racing toward autonomous security operations. GTT uses automated response internally as its first customer. Its experience over the last six months showed that many organizations are moving too fast. A poorly executed automated response strategy can create new attack surfaces, competing bots, and insecure cloud-based data exchanges.

GTT's model draws a line at the work done. Humans make decisions, and AI carries out actions once a decision has been made. The company says it learned from its own internal development and is committed to human oversight. That is the correct call. The rush to autonomous security operations agents is outpacing the governance needed to keep them safe. Showing operators the raw data alongside the AI's reasoning will build the trust that customers will need when they are ready for more automation.

The response module still matters. Agentic runbooks can reduce the time between a finding and an action. But the governance around those runbooks is the differentiator. If a customer cannot see why the system recommended an action, or cannot limit what the system is allowed to execute, then speed becomes a risk rather than a benefit.

Where the questions remain

GTT is positioning Defense Halo to unify network and security operations. The logic holds. An anomaly might be operational or malicious, and the same data should address both. But large enterprises may not be ready to merge their network and security operations centers. The org charts and budgets are deeply entrenched. GTT acknowledges that its sweet spot is mid-to-large, distributed enterprises, where convergence may be easier.

The go-to-market approach is sensible. Start with existing managed firewall and network customers and extend the platform through technology advisors. That gives GTT a base of customers who already trust it with network operations and who may be willing to test security analytics on the same estate. The real test is how the platform performs on networks messier than GTT's own, against well-funded security platforms that are moving toward network-level correlation.

There is also the question of scope. Defense Halo is not a replacement for every security tool. It is a correlation and response layer that sits close to the network. Its value depends on the quality of the data it can see and the customer's willingness to change workflows. The firewall and CVE use cases are practical entry points because they solve problems network and security teams face now, not in some future architecture.

What IT and security teams should test

  • Audit data movement costs. Know what shipping telemetry to external analytics platforms costs in terms of egress, latency, and blind spots. That is the baseline for judging on-net alternatives.
  • Start with firewall policy hygiene. If you run firewalls from multiple vendors, cross-brand policy analysis and CVE correlation are low-risk, high-return places to test AI-driven security. Ask for a proof of concept on a real portion of your estate.
  • Insist on isolation. In regulated industries, make customer-isolated models and in-region processing requirements explicit.
  • Separate decisions from actions. Whatever AI security platform you use, set policy on what AI can recommend, what it can execute, and what always requires human sign-off. Be skeptical of vendors that cannot show their reasoning alongside the raw data.
  • Converge the data before the teams. Give network and security operations staff a shared, correlated view first, and let organizational changes follow the results.
  • Pressure-test consolidation claims. Model licensing, integration, and headcount savings using your own numbers, not the vendor's case study.

The near-term test is how Defense Halo performs on networks that are messier than GTT's own and against larger security platforms that are moving toward network-level correlation. GTT's stated sweet spot is mid-to-large distributed enterprises, starting with existing managed firewall and network customers and extending through technology advisors.


Source: Network World News


Share:

Your experience on this site will be improved by allowing cookies Cookie Policy