Foremy Software Insider Report — a weekly look inside the tools, deals, and decisions shaping the software industry.
The pattern behind the headlines
Every few weeks brings a new breach disclosure, a new CVE with a memorable name, a new round of “what should organizations do differently” commentary. It’s easy to treat each incident as an isolated event — a single vendor’s failure, a single team’s oversight. But insiders who work across incident response, security architecture, and vendor risk management increasingly describe this year’s breach activity as symptomatic of a structural gap rather than a string of unrelated bad luck.
The gap, in short: most enterprise security programs were designed around a network perimeter that no longer meaningfully exists, while most enterprise software procurement was designed around a trust model — “we vetted this vendor once, therefore we trust everything it touches indefinitely” — that modern software supply chains have quietly outgrown.
Three failure patterns repeating across incidents
1. Third-party and supply-chain exposure
A large share of this year’s most damaging incidents did not originate from a direct attack on the victim organization’s own infrastructure. They originated through a vendor, a plugin, an integration, or a piece of shared tooling that had broader access than the situation actually required. This is not a new category of risk — supply-chain compromise has been discussed for years — but the scale of interconnected software dependencies in a typical enterprise stack has grown faster than most organizations’ ability to inventory, let alone monitor, those dependencies.
Security leaders increasingly describe the core problem as a visibility gap: most organizations genuinely do not have an accurate, current list of every third-party system that has some form of access to their data, let alone a risk rating for each one.
2. Credential and identity compromise, not “hacking” in the cinematic sense
The popular image of a breach still involves some form of sophisticated exploit against a firewall. The far more common reality this year, according to incident responders, is far less dramatic: a stolen credential, a phished employee, a session token lifted from a compromised personal device, or an over-permissioned service account that nobody had reviewed in years. Identity, not the network edge, is where most of this year’s real damage originated.
This is precisely the thesis behind zero-trust architecture — the idea that no user, device, or service should be implicitly trusted based on network location alone, and that every request should be continuously verified. Zero-trust has been an industry buzzword for close to a decade, but this year’s incident data is giving security budget owners a much easier argument for actually funding the unglamorous identity and access management work that zero-trust requires, rather than treating it as a slide-deck aspiration.
3. Slow detection, not just initial compromise
A recurring, uncomfortable theme across post-incident reviews this year: the gap between initial compromise and detection remains uncomfortably long at many organizations, even ones with mature-looking security tooling. Attackers who gain a foothold frequently have substantially more time to move laterally, escalate privileges, and locate valuable data than security teams would like to admit publicly. Detection speed, not just prevention, is emerging as a primary axis that boards are starting to ask about directly.
“Boards used to ask ‘can this happen to us.’ Increasingly they’re asking ‘how long would it take us to notice if it already had.’ That’s a healthier and much harder question.” — a CISO at a large logistics company, speaking on background
Where security budgets are actually shifting
Talking to security leaders and vendors across the ecosystem this month, a few clear budget-priority shifts stand out relative to a year or two ago:
- Identity and access management is absorbing a growing share of security spend, often at the expense of legacy perimeter tools that organizations are quietly deprioritizing.
- Third-party risk management platforms — tools that continuously monitor vendor security posture rather than relying on a once-a-year questionnaire — are seeing renewed enterprise interest after years of being treated as a compliance checkbox.
- Detection and response tooling is being evaluated less on feature checklists and more on measurable time-to-detect and time-to-contain benchmarks, with organizations increasingly running their own tabletop exercises to test vendor claims rather than taking them at face value.
- AI-specific security concerns are a genuinely new budget line for many organizations — governance around what internal data AI tools and agents are permitted to access, and monitoring for a new category of prompt-injection and data-exfiltration risks that didn’t exist as a meaningful attack surface a few years ago.
The AI security wrinkle
The rapid enterprise adoption of AI assistants and agentic tools (see our companion piece on AI coding tools) has quietly introduced a new and still poorly understood attack surface. When an AI agent has been granted access to internal documents, codebases, or systems in order to be useful, it inherits a version of the access-control problem that security teams already struggle with for human employees — except with far less industry experience in how to monitor and constrain it. Several security researchers this year have demonstrated that carefully crafted content — a malicious instruction embedded in a document, an email, or a webpage — can manipulate an AI agent into taking unintended actions if the agent isn’t properly sandboxed and permissioned.
This is prompting a new category of tooling focused specifically on AI agent governance: monitoring what data and systems agents actually touch, constraining their permissions to the minimum necessary for a given task, and building audit trails that let security teams reconstruct what an agent did and why, after the fact.
Practical takeaways for security and IT leaders
- Build (or rebuild) an accurate third-party inventory. This sounds basic, but insiders consistently describe it as the single most impactful and most neglected step. You cannot manage risk from systems you don’t know you’re exposed to.
- Prioritize identity hygiene over new tooling purchases. Multi-factor authentication coverage, service account review, and privilege minimization deliver outsized risk reduction relative to their cost, and don’t require a major new procurement cycle.
- Treat AI agent permissions with the same rigor as employee offboarding. Agents should have narrowly scoped, auditable access — not broad standing permissions granted once during setup and never revisited.
- Invest in detection speed, not just prevention. Given that breach prevention will never be perfect, the organizations that fare best are consistently the ones that detect and contain fastest, not the ones that never experience an incident at all.
What to watch next
- Whether regulators move toward mandatory, standardized disclosure timelines for AI-agent-related incidents specifically, distinct from traditional data breach disclosure rules.
- Whether cyber-insurance underwriters begin pricing AI agent deployment as a distinct risk factor, the way they already price cloud misconfiguration risk.
- Whether this year’s identity-focused budget shift proves durable, or reverts once incident headlines fade from board attention, as has happened with past security spending cycles.
A quick insider Q&A
Q: Is zero-trust architecture actually achievable, or is it mostly a marketing term at this point?
A: Security architects we spoke with are split, but the more experienced ones tend to describe zero-trust less as a binary state you achieve and more as a direction you continuously move toward. Very few organizations can honestly claim to have eliminated implicit trust everywhere across a large, legacy-heavy environment. What separates organizations making real progress from ones just adopting the marketing language is whether they can point to measurable reductions in standing access and broad service-account permissions, not whether they’ve purchased a product labeled “zero-trust.”
Q: Should smaller companies with limited security budgets prioritize identity work over other investments?
A: Most incident responders we talked to say yes, specifically because identity hygiene improvements are disproportionately cheap relative to the risk they reduce. Enforcing multi-factor authentication broadly, auditing and pruning service account permissions, and eliminating standing administrative access typically cost far less than a single new security product purchase, yet address the root cause behind a large share of this year’s actual incidents.
Q: How worried should organizations be, realistically, about AI agent-specific attacks right now?
A: Researchers describe the current risk as real but still mostly theoretical at enterprise scale — proven in controlled research settings and isolated incidents, not yet a dominant category of actual breaches. The bigger near-term risk, several noted, is less about sophisticated attacks and more about organizations granting AI agents broad, poorly audited access simply because it was the path of least resistance during setup, creating exposure that may not be exploited immediately but sits there as unmanaged risk.
A note on incident disclosure culture
One quieter shift worth flagging: the tone of public breach disclosures has changed noticeably over the past year. Where early disclosures in the industry tended toward minimization and carefully lawyered language, a growing number of organizations — partly under regulatory pressure, partly under competitive pressure to appear transparent — are publishing more detailed post-incident technical write-ups. Security researchers broadly welcome this trend, arguing that the entire industry learns faster when root-cause detail is shared rather than buried in vague statements about “unauthorized access.”
The bottom line
This year’s breach wave isn’t primarily a story about more sophisticated attackers. It’s a story about defensive architecture that hasn’t fully caught up to how modern software is actually built, connected, and increasingly operated by autonomous agents rather than humans alone. The organizations treating this as an opportunity to fix the underlying identity and visibility gaps — rather than simply buying another point-solution to fend off the specific attack in this week’s headline — are the ones insiders expect to be meaningfully better positioned a year from now.
Foremy’s Software Insider Report publishes weekly analysis on the tools, deals, and decisions shaping enterprise and developer software. Have a tip or an inside perspective to share? Reach the team at team@foremy.com.