The supply chain doesn’t break at logistics anymore. It breaks at login.

According to the Thales Digital Trust Index 2026 report, 89% of partner users have delayed or abandoned work due to access issues. Let that sink in and imagine the impact on your topline.

This elevates the issue from an IT convenience to revenue at risk, forcing organizations to examine how friction in their partner/supplier/reseller onboarding and authentication processes might be undermining bigger, more long-term gains. 

Smooth Partner Onboarding is the Exception, Not the Rule – Unfortunately

The Thales report reveals that among partners, only 22% receive working login details immediately; the rest have to wait.

In addition:

• Only 30% get full permissions the first time
• 66% lose access to external partner systems while they still need it
• 92% experienced access issues overall

The problem (beyond what’s already been laid out) is that in these situations, partners find workarounds. When faced with access friction, nearly two-thirds (66%) have shared or borrowed credentials, with over half citing slow official processes as the cause. 

These issues present even more business risks. Not only does delayed access prevent partners from doing their work or delivering value, but it also leads to the kind of technical fence-jumping that could cause companies to lose value they’ve already gained. 

Beyond the productivity hit, delayed access and credential workarounds accumulate what the report calls ‘silent security debt’, a growing exposure to breaches and compliance risk that rarely gets priced into onboarding cost calculations.

When it takes an average of 30-60 days for third-party assessments alone and 1-6 months to bring them on board, these kinds of delays are as expensive as they are inexcusable. 

Putting a Price Tag on Onboarding Delays

If your partners aren’t logging in cleanly and immediately, they’re not fully onboarded. And if they’re not fully onboarded, you’re not seeing full ROI.

Over a quarter (26%) of companies report losing $500k or more due to partner integration issues. Slow onboarding leads to missed deals, delayed time-to-revenue, and partner churn, incurring additional costs along the way. 

It costs thousands of dollars to onboard a supplier, and companies are looking to make that up with quick time-to-value. Login issues, false starts, and ongoing friction not only delay start times but also throw off long-term timelines, wreck projected productivity, and force teams to redo expectations. The result: work waste and revenue loss.

To highlight just one area of loss, password resets still cost around $70 per incident, and, applied to real user populations, that equates to $37,800 annually for 3,000 registered users. Whether those users are your employees or third-party employees using those credentials to log in to your systems. 

It is worth noting that most IAM administration is still done in-house, but research shows that more partners are willing to take this on. More on this in the next section. 

In the worst cases, too much friction could lead to missed opportunities, and this holds true across the board. A quarter of businesses report losing a deal due to slow legal onboarding processes. According to Fenergo’s 2025 Financial Crime Industry Trends report, 70% of financial institutions globally lost clients due to slow or inefficient onboarding. 

And like the opening stat introduced, “89% of partner users have abandoned or delayed a work task with an external partner in the past 12 months due to a website or app issue.” No matter the industry, or if it’s B2B or B2C, IAM friction in the onboarding process is a cost center, but it’s one that can be avoided.

What Partners Want out of B2B Access?

The Thales research revealed an interesting trend that could help companies offset the cost of B2B access: shifting more of the burden to partners themselves. Perhaps not surprisingly, partners want that. 

According to the report, partners want more self-service access, so access is increasingly in their hands. Over half (54%) wanted the ability to reset authentication factors themselves, and 52% wanted to see their current entitlements.

Delegated User Management, a key component of Extended Enterprise Access, fills this need, allowing organizations to maintain overarching control over identity systems while the responsibility for third-party access is offloaded to the third parties themselves:

The host organization decides which IAM controls can be delegated. A delegated manager within the partner organization takes the lead on that, acting as a proxy admin for their users. That partner has the power to grant, revoke, or update access for their team, and all requests go through them. 

This leads to offset costs for the host while resulting in greater (and faster) ROI. IT gets fewer repetitive tickets; access issues get resolved faster, so teams can get back to work sooner; and third-party IAM issues don’t result in greater costs as the partnership scales. 

Offloading IAM may be step one. But no matter where the onus resides, automation has to be step two. 

Making Automated B2B Access the Norm

When surveyed by Thales, only 22% reported that system access was automated or “provided immediately.” That means four out of five had to wait. And in an era of AI and hyperautomation, many are wondering why. 

The absence of these automated workflows takes a toll in other areas as well: only 19% reported that access changes keep pace with working needs. This can be a problem. A sale could be lost if the distributor can’t retrieve the proper inventory information. Or a competing vendor may be quoted when a reseller can’t access pricing tools.

Partner/B2B IAM is a Financial Issue

All this underscores the fact that B2B IAM extends far beyond convenience and PX. IAM failures delay projects and billing cycles. They propagate outdated information and force host companies to bear supplier costs they shouldn’t.

This makes partner IAM not just an administrative issue, but a financial one with revenue at stake. 

Because people are accessing these systems, friction leads to frustration, errors, time loss, and a domino effect of lost productivity. It sets everything off on the wrong foot, and in the case of many deals, things won’t recover. 

Partnerships don’t have to come with a buffer for operational drain. To ensure they don’t, senior leaders have to take B2B access out of the IT weeds and into the boardroom. 

INTERESTING POSTS

In this post, I will show you how I’d use the GLM-5.2 API for a long-running coding agent.

If I were setting up a coding agent today — the kind that’s supposed to run for hours, work through a real codebase, and come back with something actually finished rather than a half-baked first attempt — GLM-5.2 api is where I’d start.

Not because it’s the flashiest name in the room, but because long-running agents fail in a very specific way, and this is one of the few models actually built with that failure mode in mind.

The Problem Nobody Warns You About Until It Happens

Here’s what nobody tells you before you build your first long-running agent: the failure isn’t usually a dramatic crash. It’s slower and quieter than that. The agent starts strong, makes good decisions for the first stretch of a task, and then somewhere around hour two or three, it starts losing the plot.

It forgets a constraint it agreed to earlier. It re-does work it already finished. It confidently “fixes” something that wasn’t broken. None of this shows up in a quick demo, because demos are short. It only shows up once you actually let the thing run.

That’s the exact scenario GLM-5.2 was built around, which is why it’s the model I’d reach for first rather than whatever’s topping a general leaderboard that week.

Why I’d Trust It to Actually Stay on Track

What sold me isn’t a single benchmark number, it’s the pattern across all of them. GLM-5.2 keeps showing up strongest specifically on the evaluations that simulate messy, hours-long engineering work rather than quick, self-contained problems.

On the kind of test that mimics an agent looping through a real repository — reading, planning, executing, checking its own work — it holds up close to the very best models out there, closed or open. That’s a different skill than being clever on a single tricky question, and it’s the one that actually matters once your agent’s task list is measured in hours instead of minutes.

Giving It Room to Actually See the Whole Project

The other piece I’d lean on hard is how much it can hold in view at once. A long-running agent working across a real codebase needs to see more than the one file it’s currently editing — it needs the surrounding context, the conventions the rest of the project follows, the pieces it touched three steps ago. GLM-5.2’s context window is big enough to hold a genuinely sizeable project in view all at once, instead of forcing the agent to work blind, one small slice at a time, and hope nothing important got left out of frame.

That’s less “nice to have” and more the actual difference between an agent that understands what it’s working inside of and one that’s just guessing.

If you’re building this kind of setup yourself, it’s worth spending a little time in the GLM-5.2 api listing before committing to anything — seeing what it sits next to, and what the nearby options trade off, tends to clarify what you actually need a lot faster than reading a single spec sheet in isolation.

The One Feature I Didn’t Expect to Care About This Much

GLM-5.2 can also be paired with a web-search tool during a task. Instead of switching to a separate model version, the agent can invoke search when it needs current documentation, recent release notes, or an external source to verify an assumption. That sounds like a small add-on until you picture what a long agent run actually needs.

A coding agent working for hours doesn’t just need to reason about the code in front of it — it occasionally needs to check something outside its own head. Did a library’s API change since the model’s training data was collected? Is there a known issue with the exact error it just hit? Rather than the agent confidently guessing at an answer that might be out of date, it can actually go look, mid-task, without a separate setup. For anything that touches fast-moving dependencies or recent library changes, that’s the difference between an agent that occasionally hallucinates its way past a stale assumption and one that just checks. 

Where I’d Actually Draw the Line

I wouldn’t pretend this is the right tool for everything, and I’d be doing you a disservice if I acted like it was. If the job is a tight, well-specified algorithmic problem — something closer to a competitive programming puzzle than an open-ended engineering task — there are other models built specifically to be faster and cheaper at exactly that kind of question, and I’d route to one of those instead rather than defaulting to the same model for every job.

The honest way to think about it is that GLM-5.2 is the specialist for staying coherent across a long, messy, multi-step project, not the fastest or cheapest option for every single request an agent might make along the way. A GLM-5.2 API for coding comparison against a more math-and-algorithms-focused alternative is worth reading before you commit your whole pipeline to one model, because the right answer genuinely depends on which of those two failure modes worries you more.

What I’d Actually Build

If I were setting this up for real, I wouldn’t hand every single request to the same configuration. I’d let the long-horizon planning and the bulk of the repository work run through GLM-5.2, with the web-search variant switched on specifically for the moments the agent needs to check something outside its own training — a library version, a recent breaking change, a fix someone already published.

And I’d keep a faster, narrower model on standby for the quick, tightly scoped questions that don’t need any of that context or reasoning depth. That’s not a compromise so much as it’s just matching the tool to the job, one more time, the same way I’d want any agent I trust with real work to be built.

INTERESTING POSTS

In this post, I will be talking about protecting business data when teams work from different locations.

The transition to remote work has shifted the approach companies use for safeguarding sensitive data. When staff members access company systems from geographically dispersed locations, traditional perimeter defences lose effectiveness, creating new entry points that adversaries seek to exploit. Threat actors frequently target remote desktop connections as these often present vulnerabilities for business networks.

Since 2020, there has been an increase in cyberattacks aimed at remote access systems. Attackers actively seek opportunities to breach business defences when individuals operate from home offices or off-site locations. Insufficient security protocols in remote desktop deployments open the door to malware, ransomware, or data theft. Lapses such as use of weak passwords or delays in applying necessary updates further raise the risk.

For organisations managing teams across different sites, maintaining a balance between data accessibility and strong protection is now essential. Secure remote access depends on layered measures that include strong authentication, full encryption, ongoing monitoring, and regular user security training. Companies adopting this strategy help sustain productivity and resilience while making severe security breaches less likely.

The Rising Security Risks of Remote Work Environments

The rapid shift to remote work has created increased opportunities for cybersecurity threats. Many businesses now support remote work, but a significant number lack proper security measures. Applying strong cyber hygiene and choosing to work securely with remote access software that builds in both data encryption and strong user authentication can close common gaps. This approach ensures that endpoints and connections benefit from layered protection, lowering risk from exploitable weaknesses.

Data breaches related to remote work vulnerabilities have become a growing concern for businesses, with costs and risks continuing to rise as more organizations adopt remote and hybrid work models.

Attackers frequently target Remote Desktop Protocol (RDP) connections. Common attack tactics include brute force password attempts, man-in-the-middle interceptions, and abusing unpatched RDP vulnerabilities. Traditional VPNs often fail to address all security risks in remote work settings because their protection usually ends at securing the tunneling of data between endpoints.

Essential Security Measures for Remote Desktop Access

Strong authentication forms the foundation of secure remote access. Multi-factor authentication (MFA) should be mandatory for all remote connections. This combines something users know (password), something they have (mobile device), and sometimes something they are (biometrics). MFA is widely recommended by security experts to reduce unauthorised access risks.

Data encryption protects information during transmission. All remote desktop connections should use end-to-end encryption with TLS 1.2 or higher because unencrypted sessions give attackers an opportunity to view or hijack sensitive data. Session management policies add another security layer. Organisations should implement automatic timeouts that disconnect idle sessions after a set period to prevent unauthorised access if users leave devices unattended.

IP filtering and geolocation restrictions limit connection sources. Companies can configure remote access systems to accept connections only from approved locations or IP ranges, blocking connection attempts from high-risk regions or unexpected places.

Implementing Zero Trust Architecture

Zero trust architecture requires companies to verify every user and device at every step, regardless of where the connection originates. For implementation, organisations should begin with segmenting their network so remote desktop users only access the exact resources needed for their tasks, and nothing more. Instead of full permissions, IT teams can assign application-level access and apply controls that check identities continuously, not just at login.

Administrators need to use strong authentication for every access attempt, apply end-to-end encryption for all remote sessions, and monitor user behaviour in real time. If any unusual activity is detected, such as a login from an unfamiliar location or attempts to access restricted files, system settings should automatically block the session and alert security staff.

Continuous verification is a key part of zero trust, unlike traditional perimeter-based security that validates users only at initial login. Remote desktop software with zero trust capabilities constantly monitors sessions for suspicious activities and can terminate connections if unusual patterns emerge. Using least privilege access is a practical step toward zero trust. Organisations should restrict user permissions to only what is necessary for their specific job functions.

READ ALSO: The Role of Scalable Hosting in Remote Work Environments

Data Protection Strategies Across Distributed Teams

Organisations must carefully plan their data storage methods for remote teams, because each approach creates different security and management needs. Centralised storage means all business data stays on company-controlled servers, making it easier for IT teams to enforce uniform security policies, back up files, and maintain compliance. 

However, securing remote access to these centralised resources becomes a major concern, as attackers may target remote desktop gateways or VPN connections searching for a single point of entry. Companies therefore need to implement identity verification, encrypted tunnels, and regular access audits to prevent breaches.

With distributed storage, some data is kept directly on remote user devices to speed up workflow and improve offline access. While this convenience allows staff to access documents without lag or connectivity barriers, it puts a heavier security burden on endpoint devices. Organisations need strict endpoint security protocols, including required encryption and automatic update enforcement.

Data classification systems help manage information security across locations. Companies should categorise data based on sensitivity (public, internal, confidential, restricted) and apply appropriate security controls to each level. This helps ensure the most sensitive information receives the strongest protections.

READ ALSO: How to Compare VPN Services Without Falling for Marketing Claims

Creating an Effective Remote Work Security Policy

A strong remote work security policy needs several key components. These include device requirements, network security standards, authentication protocols, data handling rules, incident reporting procedures, and acceptable use guidelines. The policy should be clear, accessible, and regularly updated to address emerging threats.

Employee training is important for policy effectiveness. Each member of a remote workforce should regularly take part in security awareness training. This covers how to create strong passwords, ways to identify phishing attempts, methods for sharing files securely, and steps for setting up a home Wi-Fi network to reduce outside risks.

Incident response planning must cover remote access breaches. For this, companies should build clear procedures focused on remote work realities. Organisations should set up a central reporting channel like a dedicated email or hotline so staff can immediately flag suspicious activity including unauthorised logins or malware alerts.

Recovery plans should include actions like resetting passwords, restoring files from secure backups, and providing guidance to affected users about next steps. Clear communication protocols help everyone know what information must be shared and with whom. Regular practice drills help strengthen the process and support rapid, coordinated action during real threats.

Some industries are subject to additional legal and regulatory requirements regarding data protection and remote access. Organisations operating in these sectors should ensure their remote work solutions and policies align with relevant standards and undergo regular security assessments to identify and address potential vulnerabilities.

INTERESTING POSTS

Discover how AI models are revolutionizing cybersecurity workflows through automated threat detection, incident response, secure development, and intelligent security operations.

Cybersecurity has always been a field where speed matters a lot. That’s because attackers are fast and new vulnerabilities come up almost every day. The timeframe between a threat appearing and it causing damage is sometimes only a few minutes.

Over the years, security teams have been fighting this battle with tools and processes that weren’t built for the speed of modern threats. Now, AI models are offering a much-needed change. 

AI is transforming how cybersecurity works in threat detection, incident response, and vulnerability management. It’s not by replacing the humans already doing it. Instead, AI provides capabilities that weren’t possible before, and we’ll discuss how as you continue reading.

The Major Problem AI Is Solving in Cybersecurity

Before we go into why AI is having such a significant impact on cybersecurity workflows, let’s examine the problem. In one word, it’s volume.

Modern organisations generate large amounts of data, including network logs, authentication events, application behaviour, and more. Going through all that data manually to find information is not only difficult but more or less impossible at the speed threats demand.

Traditional security tools rely on rule-based systems. In other words, they use predefined conditions that trigger alerts when specific patterns are detected. These systems work well for known threats, but they struggle with newer attack methods. As a result, they often generate many false positives and require constant manual updating to stay relevant.

AI models fix the problem by learning from data rather than following fixed rules. They identify patterns, adjust to new behaviours, and improve over time. That’s why they’re a better fit for modern cyber threats.

Detecting Threat and Analysing Behaviours

An essential application of AI in cybersecurity is to spot threats early. Rather than looking for specific known attack patterns, AI models recognise normal behaviour for users, devices, and network traffic. Then, they flag significant differences once detected. This approach stops threats that rule-based systems miss entirely. 

Security teams using AI-powered behavioural analysis achieve two things. First, they report significant reductions in the volume of alerts they have to review. Second, they spend less time identifying threats, meaning more focus on real incidents.

Accessing AI Models for Cybersecurity Applications

Based on our research, a significant development in how organisations use AI for cybersecurity is the rise of flexible model platforms. Teams don’t have to build proprietary AI systems from scratch, which can be expensive and time-consuming. They can instead access powerful AI models through APIs. 

One of the best platforms for that is API QIK, which connects to many leading Large Language Models (LLMs). By providing a single access point to many AI systems, it makes a rather complex job much simpler. Plus, the platform helps reduce costs because you work with the best model per task.

For teams evaluating options, a unified AI model API platform is effective. It allows security engineers to test, compare, and deploy different models for various use cases. Such flexibility in a single platform significantly reduces the time from creating a concept to deploying it.

READ ALSO: What is the Digital Personal Data Protection Act and Who Does It Apply To?

Improving Incident Response

When an incident happens, how fast cybersecurity teams respond is crucial. The longer an attacker has access to a system, the more damage they can do. In this area, AI models are reducing the duration from detection to containment in ways that manual processes just can’t match.

With an AI-assisted incident response tool, it’s possible to correlate events on multiple systems. This way, teams can reconstruct an attack timeline, identify affected assets, and recommend actions to take. All this happens within seconds of a threat being detected.

To be clear, the automated process doesn’t remove the need for human judgment. At the end of the day, experienced analysts still make calls on how to respond. However, AI handles the time-consuming groundwork, so security professionals can focus on decision-making.

Large-Scale Vulnerability Management

Identifying and patching vulnerabilities before attackers can exploit them is one of the most important tasks in cybersecurity. Not to mention, it’s resource-intensive. Most organisations deal with more vulnerabilities at a time than they have the capacity to fix, so they’re forced to prioritise.

AI models help bring intelligence to this process. They don’t just list every known vulnerability by severity score. Instead, they can assess the actual risk each vulnerability poses, based on what the organisation does. As such, it’s a kind of contextual prioritisation that helps security teams channel resources toward the vulnerabilities that matter most.

AI Models and the Security Operations Centre

The Security Operations Centre (SOC) is where much of the day-to-day work of cybersecurity happens. You have analysts monitoring alerts, investigating suspicious activity, and responding to incidents. Basically, they work to stay ahead, so it’s demanding.

With a unified AI model marketplace in the SOC workflow, teams can skip the routine, repetitive tasks. We’re talking about actions such as log analysis, alert triage, and threat classification. The organisations making the most of this are those with a clear workflow for how AI outputs feed into human decision-making.

Human Review Remains Necessary

At this point, we want to be clear about something: AI is not replacing cybersecurity professionals. The industry threats are too complex and dependent on contextual judgment for that to happen.

What AI is doing is changing what cybersecurity professionals spend their time on. The systems allow less time sorting through noise or performing manual and repetitive investigation work. That leaves more room for strategic decisions, complex threat analysis, and the kind of creative thinking that attackers themselves use.

FAQ

Can AI replace cybersecurity professionals?

No. AI enhances productivity by automating repetitive tasks, but human expertise remains essential for investigation, decision-making, and strategic planning.

What cybersecurity tasks benefit most from AI?

Threat detection, log analysis, malware analysis, incident response, vulnerability assessment, secure coding, and threat intelligence processing are among the areas seeing the greatest impact.

Why use multiple AI models in cybersecurity?

Different AI models excel at different tasks. Combining them allows organizations to leverage the strengths of each model for more accurate and efficient workflows.

Bottom Line

The current trend of AI in cybersecurity isn’t temporary. From all indications, AI is becoming fundamental to how security work is done across the industry. It applies to detection, response, vulnerability management, and threat intelligence. 

Organisations that prioritise AI integration today will be better positioned to handle the threats of tomorrow. The good news is that the tools are more accessible than ever. Also, there are well-proven use cases, and the operational advantages are real. So, in conclusion, AI matters in cybersecurity, and more organisations need to build it into their workflows.

INTERESTING POSTS