Clement Brako Akomea Table of Contents
Modulate has secured $25 million in new deepfake detection funding, SecurityWeek reported on September 28, 2026. Future Ventures led the round, with Hyperplane and Lakestar taking part.
The company builds artificial intelligence that listens to human conversation and flags synthetic speech. The investment lifts its total funding to $60 million.
Modulate plans to spend the money on hiring, infrastructure and developer tools, as voice becomes a main way people interact with AI systems.
Deepfake Detection Funding: Key Takeaway
- Modulate raised $25 million, bringing its total to $60 million, to expand voice AI models that detect deepfakes and risky behavior in live conversations.
This section contains affiliate links; we may earn a commission at no cost to you.
Voice cloning attacks often start with a fake identity, so verification matters.
- GetTrusted: identity and trust verification that helps organizations confirm who they are dealing with.
- CyberUpgrade: cybersecurity compliance and management for teams that need clear policies and evidence of controls.
- Trainual: employee training and onboarding, useful for teaching staff how to handle suspicious calls.
What Modulate Announced
Modulate, which describes itself as a frontier AI firm, announced an additional $25 million investment from Future Ventures. Hyperplane and Lakestar also participated. The new money brings the company’s total funding to $60 million, according to SecurityWeek’s report on the round.
The company develops AI models that work directly on audio. Its stated goal is to help machines understand the nuances of human conversation, including whether a voice is real or synthetic. A deepfake is media created or altered by AI to imitate a real person, and a synthetic voice is one generated by software rather than spoken by a human.
The funding will support several areas of growth. Modulate says it will expand its team, build out its infrastructure, and widen its models, APIs, SDKs, integrations and deployment options. An API is a set of rules that lets one program request services from another, and an SDK is a toolkit that helps developers build software.
Who Backs the Round
Future Ventures is the investor named in connection with the additional $25 million. Hyperplane and Lakestar joined the round. The source does not disclose the valuation, the split between investors or the earlier funding stages that make up the remaining $35 million of the total.
Why the Timing Matters
Voice is becoming a main interface for AI products, from customer service agents to assistants that talk back. That shift gives attackers a new channel to abuse. A cloned voice can persuade an employee to share information or approve a request.
Modulate CEO Carter Huffman framed the problem in those terms. “Voice is becoming a primary interface for AI, and that creates a whole new set of problems that can’t be solved from a transcript,” Huffman said.
How Modulate’s Voice AI Works
Most voice analysis tools convert speech to text first and then analyze the words. Huffman argues that this approach loses information. Tone, emotion and the manner of speaking do not survive transcription, and those signals often reveal manipulation or synthetic speech.
Modulate instead builds its models around the sound itself. This design lets the system examine how something is said, and not only what is said.
The Ensemble Listening Model Architecture
The company’s technology relies on a proprietary design called the Ensemble Listening Model, or ELM. The architecture generates specialized AI models, each built for a narrow listening task. An ensemble is a group of models whose results are combined into one judgment.
According to SecurityWeek, the ELM approach produces the models that power Modulate’s Velma platform. Modulate’s own website also describes ELM as a way to combine what is said with how it is said, and to look at many audio signals beyond the transcribed words.
The Velma Platform
Velma combines more than 100 of these models. Together they detect emotion, tone, intent, synthetic speech and conversational behavior. The platform works in real time, so an application can step in during a conversation instead of reviewing it afterward.
That timing matters for security. A fraud attempt over the phone can succeed in minutes. A tool that only flags a recording the next day helps investigators, but it does little to stop the transfer or the data leak that already happened.
Claimed Accuracy and Scale
Modulate says Velma is twice as accurate as traditional large language models, or LLMs, and produces seven times fewer false positives. A false positive is a harmless event that a security tool wrongly flags as a threat. These figures come from the company and the source does not describe how they were tested.
The company also says it analyzes more than 10 million hours of audio each month and 600 million hours cumulatively. Its transcription and deepfake detection models ranked first on public benchmarks, including Hugging Face, according to the company. Hugging Face is a platform where developers share AI models and compare their performance.
Modulate’s website gives slightly different figures than the news report. It cites more than 563 million hours of conversations, a claimed 98.9% accuracy on deepfake detection and a first place ranking on a Hugging Face deepfake detection leaderboard. The gap in hours likely reflects the date each number was published. Readers should treat all of these numbers as vendor claims until independent testing confirms them.
Where Modulate’s Technology Is Used
SecurityWeek lists several typical uses for the platform. Each one shows a different way that voice analysis can support security or safety.
Protecting Healthcare From Deepfake Attacks
Modulate lists healthcare institutions among the organizations it protects from deepfake attacks. Hospitals and insurers handle sensitive patient data and take many calls from people who must prove who they are. A convincing fake voice could help an attacker talk a help desk into resetting an account.
Monitoring Voice Agents
Companies now deploy voice agents, which are AI systems that hold spoken conversations with customers. Modulate says its tools help teams monitor whether those agents perform as intended. The platform also helps voice agents understand emotion and respond with more empathy.
Reducing Harm on Social Platforms
The company points to reducing extremism on social platforms and detecting child grooming attempts. In these cases, the tone and behavior in a live voice chat can expose risk that a text transcript would miss. Moderators cannot listen to every conversation, so automated detection can help them focus on the most serious cases.
Protecting Agents With Voice Masking
Modulate also lists protecting agents through voice masking. The source gives no detail on how this feature works. In general, voice masking changes how a voice sounds, which can shield the person speaking.
Modulate’s Position on Audio Intelligence
Huffman described what his company has already deployed. “We’re already using audio native AI to protect organizations from deepfake attacks, help voice agents understand emotion and respond with more empathy, identify dangerous behavior in online conversations, and monitor whether voice agents are actually performing the way they’re supposed to,” he said.
He also spoke to developers directly. “Developers shouldn’t have to rebuild the audio intelligence layer every time they create a new voice experience,” Huffman said. That remark points to the company’s strategy: offer a shared layer that other builders can add to their own voice products.
Modulate’s website adds enterprise details. It lists integrations with Slack, Zoom, Microsoft Teams, Zendesk and Genesys. It also says it complies with ISO 27001, HIPAA, GDPR, CCPA and the EU AI Act, and that customer data is not used for training. These statements come from the vendor and were not part of the funding report.
The Wider Market for Voice Deepfake Defense
Modulate is not alone in this field. SecurityWeek previously covered isVerified, an Israeli startup spun out of the cybercrime intelligence firm Hudson Rock. It emerged from stealth on January 15, 2026, with mobile apps for Android and iOS that aim to detect AI voice impersonation in real time.
The two companies take different approaches. isVerified targets senior executives and officials, and it analyzes calls locally on the user’s device for privacy. Modulate offers a platform that developers and enterprises build into their own voice products.
Roi Carthy, CEO of isVerified, described the underlying threat plainly. “A person’s voice can be cloned by specialized gen-AI tools in seconds. Threat actors can then use these audio deepfakes to manipulate a company’s employees,” Carthy said. That warning matches the concern behind Modulate’s new funding.
Voice attacks build on older social engineering tactics. Readers who want background can review CyberSecurityCue’s guide to vishing attacks and how to prevent them, which covers fraud by phone.
Implications of the Deepfake Detection Funding Trend
The round says something about where security spending is moving. Investors are backing tools that defend against AI generated attacks with AI of their own. The sections below examine what that means for defenders, for enterprises and for the industry.
Voice Becomes a Verified Channel
For years, a familiar voice on the phone served as informal proof of identity. Voice cloning weakens that assumption. If tools can copy a voice in seconds, as Carthy warns, then a call from a senior leader no longer proves anything on its own.
Organizations will need to treat voice like any other channel that requires verification. That means callback procedures, approval steps for payments and password resets, and detection tools that examine the audio. Modulate’s real time approach fits this model because it can alert staff while the call is still in progress.
Detection Must Work Inside the Conversation
Speed shapes the value of any detection tool. Fraud over voice channels succeeds because it moves fast and pressures the victim. A tool that flags a synthetic voice mid conversation can prompt an agent to pause, ask for extra verification or end the call.
The claim of seven times fewer false positives also matters. A detector that raises too many alarms trains staff to ignore it. Fewer false alerts can keep analysts and call center agents focused on real threats, provided the claim holds up in independent tests.
Vendor Claims Need Independent Testing
The accuracy figures in this story come from Modulate. Public benchmarks such as Hugging Face offer one point of comparison, but attackers change their tools quickly. A model that ranks first today may miss a newer voice cloning technique tomorrow.
Security teams evaluating any detection product should run their own tests with recordings that reflect their calls, languages and audio quality. They should also ask how often the vendor updates its models. This due diligence protects against overreliance on a single score.
Privacy and Compliance Questions
Analyzing conversations in real time raises privacy questions. A platform that listens to emotion, tone and intent handles very sensitive data. Modulate says it is compliant with rules such as GDPR and HIPAA and does not train on customer data, which are the kinds of assurances buyers will want to see documented in contracts.
Companies that deploy such tools must also tell callers how audio is analyzed. Compliance teams should review consent rules in each region where they operate.
Attackers Will Adapt
Defensive investment usually triggers a response from attackers. As detection improves, criminals can refine their cloned voices or shift to other channels, such as text messages and email. This pattern means voice defenses work best as one layer among several.
Email authentication, staff training, strong passwords and identity verification all reduce the chance that a single fake call causes damage. For more on how AI is changing fraud defense, see CyberSecurityCue’s coverage of an AI powered fraud prevention platform.
This section contains affiliate links; we may earn a commission at no cost to you.
Layer your defenses so one fake call cannot do serious damage.
- 1Password: a password manager that helps staff use unique credentials, so a stolen login does not open every door.
- EasyDMARC: email security and DMARC tools that help block spoofed messages that often accompany voice scams.
- Optery: personal data removal that reduces the public information attackers use to build convincing impersonations.
Looking Forward
Modulate’s $25 million round shows that investors see voice security as a growing need. The company will use the money to add staff, infrastructure and developer tools around its Velma platform.
The accuracy claims come from Modulate, so buyers should test them. Even so, the direction is clear: voice needs the same verification that email and passwords already receive.
Security teams should review how their staff handle calls that request sensitive actions, and consider adding checks that do not depend on how a voice sounds.
Questions Worth Answering
Who invested in Modulate’s latest round?
- Future Ventures provided the additional $25 million, with participation from Hyperplane and Lakestar.
How much has Modulate raised in total?
- Modulate’s total funding now stands at $60 million.
What does Modulate build?
- Modulate develops AI models that work on audio to understand human conversation, including detection of synthetic speech.
What is the Ensemble Listening Model?
- It is Modulate’s proprietary architecture that generates specialized AI models, which the Velma platform combines into one system.
What is Velma?
- Velma is Modulate’s platform, which combines more than 100 models to detect emotion, tone, intent, synthetic speech and conversational behavior.
Does Velma work in real time?
- Yes, the platform operates in real time, so applications can intervene during a conversation.
How accurate does Modulate say Velma is?
- The company claims it is twice as accurate as traditional LLMs and produces seven times fewer false positives.
How much audio does Modulate analyze?
- The company says it analyzes over 10 million hours of audio monthly and 600 million hours cumulatively.
What will the new funding pay for?
- It will support team growth, infrastructure and expanded models, APIs, SDKs, integrations and deployment options.
What are typical uses of the technology?
- Uses include protecting healthcare institutions from deepfakes, monitoring voice agents, reducing extremism, detecting child grooming and voice masking.
Also worth a look: IDrive for cloud backup and ransomware recovery, Tenable for vulnerability management, and Tresorit for encrypted cloud storage.
