Every security training program.
One platform.
Awareness training, red teaming, deepfake simulation, tabletop exercises, and full orchestrated campaigns. Built on OSES™, run from a single console, and generated on demand by your team.
OSES™ is our Orchestrated Social Engineering Simulation framework: synthetic voice, video, and messaging coordinated into one campaign that follows the request into the process behind it: the callback, the approval, the transfer. Describe the scenario in plain language and the platform drafts the playbook, assigns a persona, clones the voice, places the call, handles the callback, and reports the outcome against your sector baseline. Run it with us, or run it yourself.
"Kudos to your entire team. We haven't even seen the report and the whole company is talking about the risks of voice cloning. It's been a huge win for us already."
Five programs most teams buy separately
Awareness training from one vendor, red teaming from another, deepfake testing from a third, tabletops from a consultancy, and nothing that connects them. OSES™ runs all five off the same engine, the same personas, and the same reporting.
You generate it yourself, in minutes, whenever you need it. Point the builder at the playbook someone just fell for and it drafts the remediation module. Drop in a policy and it drafts the slides and the quiz. No content request, no vendor ticket, no waiting on a release cycle.
Adversarial engagements scoped to an objective rather than a click rate. Agents run the full chain autonomously, crossing control boundaries the way a real operator would, with the CLI and API for teams who script their engagements.
Cloned executive voice and likeness delivered where your people expect it, including live avatars that join a meeting and hold the pretext under questioning. Consent and scoping agreed before any likeness is built.
The full OSES™ campaign. A voicemail lands, a text references it, a video call confirms it, and the chain runs through to the consequential action. This is the part a single-channel phishing test cannot reach.
Facilitated sessions on your incident process, with live avatar demonstrations for boards and executives and opt-in voice cloning so leaders hear themselves. People argue with the deepfake instead of watching a slide about one.
Where the target is the verification stack rather than the person. Synthetic faces, documents, and voiceprints submitted through liveness, document checks, video KYC, and voice biometrics. See deepfake penetration testing.
What OSES™ is
An Orchestrated Social Engineering Simulation is a controlled, authorized exercise in which synthetic voice, video, and messaging are coordinated into a single campaign, and the full response is observed from first contact through to the consequential action. The consequential action is the point. Not the click.
OSES™ never asks whether someone noticed. It asks whether the process held. Verification does not require distinguishing real from fake, which is exactly why it is durable: a callback placed to a number from your own directory works identically whether the caller was synthetic, a human impostor, or entirely legitimate. Read the full OSES™ framework.
Three layers that make an OSES™ campaign real
Everything else on this page runs on these. Take one away and you are back to sending mail and counting clicks.
Built around your organization
The engine works from who your executives actually are, which approval chains carry real money or access, and which pretexts would survive scrutiny inside your company. A generic lure tests whether people are suspicious of strangers. That is not the question worth answering.
Channels that hand off to each other
A voicemail lands, a text references it, a video call confirms it. Each step borrows credibility from the last. Testing one channel at a time misses the thing that makes modern social engineering work, which is the sequence itself.
Agents that hold the conversation
Voice and avatar agents run live. They place the call, converse if answered, leave a callback voicemail if not, and handle the inbound callback on their own. No operator waits on the line, which is what lets one campaign reach an entire population.
What the team running it actually sees
Every part of a campaign is editable, repeatable, and scriptable, starting with the training your team generates itself. Nothing is a black box you have to file a ticket to change.
This is the part worth pausing on. Every other vendor makes you file a request and wait for a content team to build a module, then ship it a quarter later against a threat that has moved on. Here your own team opens the builder, picks a starting point, and has a finished course with a knowledge check the same afternoon. No content request, no professional services line item, no release cycle.
- Generate from a playbook: someone acted on a simulation, and the module writes itself around what they missed
- Create from a document: drop in a policy, AUP, or code of conduct and get slides and a quiz back
- Start from scratch on a blank canvas, or open a template and edit it
- Role-specific content for finance, HR, IT, and executives, built by whoever knows those teams
- SCORM export straight into your existing LMS, branded to your organization
Your findings, your policies, your executives, your brand. Built by your team, not requested from ours.
This is where a deepfake stops being a demo and becomes a repeatable test. A persona carries identity, role, personality, appearance, and voice as one object: a live video avatar for meetings and a cloned voice for the phone. Build the executive once, then run them across every campaign, channel, and tenant without rebuilding anything.
- Named personas with role, backstory, and authority level
- Avatar selection for live Teams, Zoom, and Meet sessions
- Voice cloning from short public samples, or a synthetic voice from the library
- Accent, delivery pace, and objection handling defined per persona
- Saved library, reusable across campaigns and tenants
Consent and scoping are agreed before any likeness is built.
Dispatch research against a person or a company, anchored on ground truth so a namesake never becomes a finding. The output feeds scenario design directly, and every source is recorded so each pretext is traceable back to public material.
- Organization, executive, and reporting-line mapping
- Identity anchoring on legal name, domain, and industry to prevent wrong-entity matches
- Public voice and video source material discovery
- Vendor, tooling, and process signals pulled from public content
- Per-target enrichment, dispatched live or queued for the periodic sweep
A click rate on its own means nothing until you know what comparable organizations produce under comparable pressure. Reporting breaks results out by channel and compares each one to your industry baseline, so an outlier that needs attention this quarter is distinguishable from a median that is itself unacceptable.
- Delivery, open, click, reply, submit, and susceptibility measured per vector
- Comparison against your own vertical rather than a global average
- Filter by sub-customer, vector, playbook, and date range, with saved views
- OSINT exposure, susceptibility, trends, and scheduled report tabs
- Export to CSV, plus board and executive briefing formats
Reporting is organizational. No named individuals, no department leaderboards.
Every channel an attacker uses, one campaign
The email gateway is the one channel with mature defenses. Modern campaigns arrive somewhere else, which is why a simulation that only sends mail tells you very little.
A synthetic executive who joins the meeting and responds in real time. Not a recording. It answers questions, handles pushback, and holds the pretext when a target tries to verify mid-call.
An autonomous agent that runs the whole call cycle. It dials, converses if answered, leaves a callback voicemail if not, and answers the inbound callback itself, adapting to objections rather than reading a script. In one engagement 21.6% of the population called back and spoke to the agent unprompted.
Executive likeness built from short public samples and delivered where your people expect it, including ringless voicemail, with caller ID and sender domain under our control rather than a spoof filter's.
Written lures generated per target rather than per campaign, across email with sender infrastructure we control, Teams and Slack messages, SMS sequences, and social impersonation. Agents reply when the target writes back.
The invite is the lure and the meeting is where compliance happens. Calendar invitations from a spoofed organizer, multi-participant scenarios with a supporting cast, and meeting-to-payment or meeting-to-access sequences.
Where most orchestrated campaigns begin, and where detection performs worst. Vishing against staff and executives, helpdesk and IT support pretexts, and phone-to-SMS or phone-to-portal handoffs.
Eight things most platforms keep behind a ticket
Every medium in one console, and the objects behind them exposed as yours to edit rather than ours to hide.
Build a campaign that crosses channels without switching tools. Conversational voice and ringless voicemail, live interactive video avatars, cloned voice and deepfake video, email, SMS, Teams, Slack, chat, and calendar-based meeting scenarios.
The prompts driving every agent are yours to edit. Set the objective the agent is working toward, define tone, urgency, and escalation behaviour, and set guardrails on what an agent will never say or ask for.
Build a caller once, then reuse them across the whole program. Named personas with role, backstory, and authority level, with voice, accent, delivery pace, and objection handling assigned per persona.
The contextual layer, assembled from public sources before anything runs. Organization, executive, and reporting-line mapping, public voice and video source discovery, and vendor and process signals from public content.
Population-wide delivery, staged the way a real campaign would arrive. Bulk import and segmentation, staggered send and call windows rather than a single blast, per-channel throttling, and timezone-aware scheduling.
Target populations pulled from your directory instead of a spreadsheet. Sync users and groups from Microsoft Entra ID and target by department, role, or group membership, with populations staying current without manual re-import.
Findings and evidence, exportable rather than trapped in a dashboard. Action, verification, and report rates per campaign, full call transcripts and per-interaction outcomes, peer vertical benchmarking, and board-ready briefing formats.
Everything the console does is available programmatically. Create, launch, and monitor campaigns via API, manage targets, personas, and prompts as data, and take webhook events for campaign activity. The CLI makes an engagement a config file.
Run it with us, or run it yourself
The first cycle typically runs fully managed so your team sees how scenarios are designed and findings are written. After that, the platform is yours to operate.
We run the simulation
- Scenario design, execution, and reporting handled end to end
- Debrief run by the team that ran the campaign
- Fully external, nothing installed
- Best first cycle for a new program
Your team builds and runs it
- Full console, API, and CLI access
- Tailoring controlled by your team, not ours
- Repeatable cycles for quarterly or annual programs
- Simulation, red team, and training bought separately
Under your brand, multi-tenant
- Many customers from one instance
- Your branding across console and reports
- Partner portal and MSSP tooling
- Bulk operations across tenants via API
Teams operating the platform
Anyone who needs orchestrated simulation to be repeatable rather than a once-a-year engagement.
Published engagements, published numbers
Both run under signed authorization, coordinated with IT, people management, and executive leadership. Client names never appear in our public material.
OSINT sourced enough public video to clone the CEO's voice. A spoofed local number rang silently to voicemail, dropped a personalized deepfake message, then followed up by SMS. The callback line was staffed by the clone, and every inbound text auto-responded with the phishing link.
A cloned voice of the local COO, driven by an agent that called each user directly and held a live conversation with anyone who answered, followed by an SMS. Everyone who clicked received a two-minute training module on the spot. No punitive measures, quarterly cadence.
After the simulation
The reaction is consistent, and it is the reason the training lands. People who have just been fooled by a synthetic voice do not need convincing that the threat is real.
Users were surprised with how good the deepfakes were. I'm really impressed. Really crazy talking to a deepfake.
I was expecting a demo, not an episode of Black Mirror. This is really good. I'm surprised at how advanced it's gotten.
The training was well-structured, clear, and provided valuable insights into the growing threat landscape associated with deepfakes.
Common questions
What is OSES™, and how is it different from awareness training?
OSES™ is Orchestrated Social Engineering Simulation: a controlled, authorized exercise where synthetic voice, video, and messaging are coordinated into one campaign and the response is observed through to the consequential action. Awareness training builds the reflex; OSES™ measures whether the reflex, the procedure behind it, and the controls around it hold under a realistic orchestrated attack. They are complementary, and training built on your own OSES™ findings is materially more effective than generic content, because the correct action is defined by your runbook.
Do we have to ask you to build training modules?
No. Your team generates them on demand, in minutes, from inside the platform. Point the builder at a playbook and it drafts a remediation module for the people who acted on that simulation; upload a policy, AUP, or code of conduct and it drafts slides and a knowledge check to refine; or start from a blank canvas or a template. Output is SCORM, so it drops into your existing LMS branded as yours. There is no content request queue and no professional services engagement to write a course.
Can we run the platform ourselves?
Yes. The first cycle typically runs fully managed so your team sees how scenarios are designed and findings are written. After that the platform is yours to operate, with self-managed, white label, and multi-tenant options for internal red teams, MSSPs, and training providers.
Do we need to install anything?
No. Engagements run fully external, with no software installed and no integration into your environment, which removes most of the change-control burden that blocks testing in regulated environments. Entra ID sync is optional and directory-only, with no agent or endpoint software.
How much of the platform is editable?
The prompt behind every agent, the objective it works toward, its tone and escalation behaviour, and the guardrails on what it will never say are all editable in the console. Personas, playbooks, and prompts are versioned and reusable across campaigns and tenants.
How does the voice agent handle callbacks?
The agent places the outbound call and converses if answered, leaves a voicemail with a callback number if not, then answers and converses on the inbound callback with no operator in the loop. Since most outbound calls go to voicemail, the callback becomes the dominant route through a population, and it inverts the trust posture completely: the target initiated contact and has already decided the counterparty is real before anyone speaks.
What scale does the voice channel support?
Up to 150 concurrent live voice sessions, with staggered call windows, per-channel throttling, and timezone-aware scheduling. There is no operator bottleneck, which is what makes population-wide voice coverage possible rather than a sampled subset.
Is there an API and a CLI?
Yes. Everything the console does is available through the REST API, and the breacher CLI drives that API from a terminal or hands it to an AI agent. Campaigns can be defined as configuration, run in CI, and their results pulled into your own tooling through webhooks and result endpoints.
How is reporting structured?
Per vector, per playbook, and per date range, with delivery, open, click, reply, submit, and susceptibility measured against your industry baseline rather than a global average. Reporting is organizational: no named individuals and no department leaderboards, because scoring individuals assigns the outcome to the layer with the least visibility into the attack.
Is consent required to clone a voice or likeness?
Yes. Consent and scoping are agreed before any likeness is built, every engagement runs under signed authorization, and generated voice and video are handled with zero retention. Sources used for the contextual layer are recorded so every pretext is traceable.
See it run against your own organization
Thirty minutes. We walk you through a real OSES™ engagement, scenario design through findings, and you decide whether your process would have held.
Or read the assessment methodology and full capability list.
support@breacher.ai
