Once, security was a defined discipline—a tight set of technical practices meant to keep systems safe. Today, it’s the junk drawer for every responsibility no one else wants to own.
The trust and safety work stripped from platforms for political expediency? Security now has to pick it up to succeed. The resilience planning deprioritized in favor of speed? Security must step in and own it. The ethical guardrails quietly written out of AI procurement rules? Security can’t ignore them because without that work, defense will fail.
The word security is doing too much work.
And the people holding the line are carrying a load no one team should sustain but if they must, they need to do it intentionally and strategically, because here’s why.
How Trust & Safety Gaps Become Security Threats
When trust and safety functions are gutted, the harms they once mitigated don’t disappear, they evolve into attack surfaces.
Gaps in content moderation, bias detection, and abuse prevention don’t just leave users vulnerable; they invite exploitation. Misinformation floods the space left by absent verification protocols. Biased algorithms become tools to disenfranchise specific communities. Without abuse reporting and enforcement mechanisms, phishing, harassment, and social engineering can run at scale without friction.
In a world of autonomous agents and AI-driven systems, understanding behavior becomes even more important. Trust and safety functions now have to apply in security contexts, or work in close collaboration with security teams, to identify anomalous or escalating agent behavior before it can be weaponized or unanticipated behavior causes unwanted harm.
Adversaries understand this better than most defenders. Erode trust, and you weaken defenses. Destabilize a community’s sense of reality, and you can operate inside the noise.
Safety failures soften the ground, security failures finish the job.
The Escalation: AI as a Combatant
Now add AI to the mix, a technology capable of scaling exploitation at machine speed, and unpatched trust gaps become accelerants, not just vulnerabilities.
As Nicole Perlroth warned in her Black Hat keynote, the signals were there: Shamoon, SolarWinds, NotPetya, Colonial Pipeline. None were Black Swans. Each built on the last. Each was a preview. And now, the warnings are blinking red.
AI is ending its honeymoon phase and moving from curiosity to combatant. It’s not just speeding up the old playbook, it’s writing a new one. Automating the kill chain. Federating initial access. Mapping critical assets. Running ransomware negotiations with maximum psychological pressure. The best APTs are already folding it into zero-day exploitation. XBOW is topping bug bounty leaderboards. Claude is winning hacking competitions. And attackers are treating autonomous agent workflows as critical infrastructure, because they are.
Combine this escalation with the absence of trust and safety guardrails, and you create an unnecessarily chaotic threat landscape. Without the friction points that once slowed attackers, verification checks, abuse detection, bias audits, offense moves faster, hides better, and exploits social fault lines with industrial precision.
Building on Brittle Foundations
Right now, we are constructing mission-critical systems on unpredictable, easily hijacked foundations. LLMs are producing secure code only half the time. That performance doesn’t improve with larger or newer models.
This is Ken Thompson’s 1983 warning, mentioned by Nicole,“You can never trust code you didn’t write yourself,” reborn for an era when we barely understand the code or the context we’re delegating to machines, including the political, economic, and social fault lines adversaries will weaponize.
And we are doing it at a pace that outstrips comprehension. Speed, the great enemy of security, is winning.
Why Security Can’t Be the Cleanup Crew
Too often, defenders are brought in late—expected to secure systems whose architecture already bakes in failure. In AI, that means patching hallucinations, prompt injection vulnerabilities, and emergent behaviors in production, in systems that were never threat modeled for autonomy or unpredictability.
This is not sustainable. You can’t firewall your way to resilience. You can’t red team your way to safety if the design itself is flawed. The longer we treat security as a bolt-on, the more brittle—and costly—our entire digital ecosystem becomes.
And when brittle systems fail at scale, accountability doesn’t just land on defenders—it cascades upward. Boards, regulators, and governments will all demand answers. The absence of comprehensive guardrails doesn’t erase responsibility; it creates a vacuum that regulators, legislators, courts, and the public will fill—usually in the wake of fresh harm, and often with outcomes far harsher than proactive design would have required.
Secure by Design with Trust Embedded for AI
Embedding security from the start isn’t a brake on innovation, it’s an accelerant. It means the AI model that ships has working guardrails. It means the infrastructure it runs on has been threat modeled for the social and political contexts adversaries will exploit. It means attribution, auditability, and behavioral constraints are part of the blueprint not rushed in after an incident.
But in today’s climate, there are moments when talking about trust—or the ethics, bias mitigation, and harm reduction work that underpin it—becomes politically inconvenient. In those cases, we need a framing that keeps trust in the build process without making it a lightning rod. That’s where Secure by Design with Trust Embedded comes in.
This approach treats trust and safety not as optional extras, but as integrated capabilities within security itself. Whether you’re accounting for AI agent behavior or anticipating the weaponization of societal context, these elements are woven into security architecture and processes. If they’re ignored, your defenses will have blind spots adversaries can and will exploit. Just as importantly, those blind spots become accountability gaps. Regulators, courts, and public opinion will eventually fill them and often with consequences far harsher than what proactive design would have required.
Secure by Design with Trust Embedded for AI means embedding security thinking from the moment an idea is conceived, not as a patch after something goes wrong. It starts with threat modeling early and often, and building in attribution and decision audit trails so actions can be traced and understood. It includes setting boundaries on cumulative agent behavior to prevent slow-drip escalation, while maintaining the safeguards that detect and respond to fairness, harm, and bias issues—signals attackers often try to remove because doing so blinds the system to entire categories of abuse. It also considers the security of inputs and outputs, the integrity of the supply chain, and the ability to monitor and adapt to model drift or changing adversary tactics. Fail-safe mechanisms and human overrides act as guardrails for the unexpected, and explainability helps ensure security-relevant outputs can be audited and improved. Above all, it acknowledges that in an AI-driven threat landscape, robust security must account for both the technical vulnerabilities and the societal contexts adversaries will exploit and that trust and safety are indispensable to doing so effectively, even when they must be integrated quietly.
By embedding trust inside security, whether or not it’s politically expedient to name it, you preserve the essential guardrails that make defense possible in the first place.
What Companies Should Do Now
The path forward isn’t about adding more checklists or compliance theater. It’s about re-architecting how organizations approach resilience from the ground up. Here are the critical shifts every company should prioritize:
Build Trust by Design or (Secure by Design with Trust Embedded): Integrate guardrails, attribution, and behavioral constraints from the earliest design stage. Treat trust and safety as core parts of resilience, not optional extras. If you still have separate trust, safety, and security teams, make sure they are working in close concert rather than as disconnected functions.
Threat Model Beyond Code: Expand risk assessments to include societal, political, and behavioral contexts adversaries will exploit. Don’t just look for buffer overflows—look for misinformation pipelines, identity exploitation, and AI-driven agent manipulation.
Invest in Continuous Oversight: Static security controls won’t cut it. Establish decision audit trails, longitudinal monitoring, and behavioral budgets for AI agents. Think less about snapshots, more about evolution over time.
Design for Accountability Across Jurisdictions: Regulatory landscapes are fractured and volatile. Aim for best-in-class governance that can withstand scrutiny anywhere, then adapt the narrative and optics to align with local expectations.
Reinforce Culture and Communication: Security isn’t just technical. Document the value of trust and security decisions, train cross-functional teams, and create feedback loops so that design, product, policy, and security functions work as one.
These steps won’t eliminate the burden defenders carry but they make it intentional, strategic, and survivable.
The Stakes
If we keep treating security as an afterthought, we’re not just risking breaches—we’re risking the viability of entire missions and business models.
Business Value: Every unmitigated vulnerability is a compounding liability, eroding trust, inflating incident costs, and inviting regulatory scrutiny. One breach can erase years of brand equity and market growth.
Mission Integrity: For public agencies, nonprofits, and critical infrastructure operators, a single compromise can derail core objectives—whether delivering public services, protecting elections, or safeguarding supply chains.
Individual Safety: In an AI-enabled threat landscape, a compromised system isn’t just a data event, it can cause real-world harm. Manipulated medical devices, hijacked autonomous vehicles, deepfake-enabled harassment.
Addressing these risks proactively doesn’t just prevent loss, it creates durable advantages. Organizations that embed trust and security from the start move faster with confidence, adapt more effectively to change, and turn resilience into a competitive edge that attracts customers, talent, and partners for the long haul.
Keep your trust, safety, and security functions robust and working in close collaboration. If you don’t have full teams in all three functions, or need to consolidate due to shifting budgets or structures, make sure trust and safety capabilities are embedded into your security foundation. This is even more critical now: recent U.S. federal procurement changes are stripping, or watering down, safety-related language in ways that conflict with state and international norms, creating a confusing patchwork of obligations. At the same time, even countries that have relaxed safety guardrails may still hold your organization accountable, especially if those gaps contribute to a security incident affecting their critical infrastructure or public trust.
The tension is already clear. Just last week, two Republican senators called for a congressional investigation into Meta’s AI policies after a Reuters article revealed its chatbots were allowed to engage in inappropriate interactions with minors. This is not a security incident—it is a content harm, a trust and safety failure—but one that could be weaponized to perpetuate security incidents. More significantly, it is also a signal that accountability will come, despite the broader Republican push for “no guardrails.” After all, guardrails aren’t a brake on innovation, they’re what make it durable.
For any jurisdiction, when the outcomes are politically indefensible or create an intolerable harm, even those who resisted regulation will demand it—and no company can afford to ignore that reality.
This regulatory confusion is only made more complex when you consider the global landscape. Between conflicting state, federal, and international requirements, the environment is fractured and volatile. The EU is moving in a very different direction from the U.S. with the AI Act, embedding trust and safety into compliance frameworks. China is charting its own path with requirements shaped by very different political and social priorities. And across the Global Majority, governments are pushing for digital sovereignty and seeking to define their own rules of the road.
For global companies, this patchwork makes one thing clear: you can’t chase every standard in isolation.
You must design for best-in-class governance—aiming for Trust by Design that fully integrates robust security. Where politics, budgets, or structures make that framing difficult, pursue Security by Design with Trust Embedded as the practical path. From there, adapt the narrative and optics to meet shifting expectations in each jurisdiction.
In this fractured regulatory environment, coupled with the reality of opportunistic malicious actors, defenders cannot hope to react. They must design for best-in-class security, Secure by Design with Trust Embedded, and build organizational structures that support resilience, alignment, and trust across every jurisdiction, no matter how politically or legally complex.
How is your organization making sure security and trust are built in from the very beginning, not bolted on after something goes wrong? And what companies have you seen doing this well?