Shigemasa Sasaki, founder of RightsTech and a commentator on digital evidence infrastructure, argues that the missing layer in trust and safety is not detection; it is preservation.
As generative tools become cheaper and easier to use, concerns about deepfake content and other forms of technology-enabled exploitation are increasing, while the systems used to document and address such material continue to develop. Sumsub’s Identity Fraud Report found that deepfake detections roughly quadrupled between 2023 and 2024 and now account for about 7 percent of fraud attempts worldwide. Victims often identify harmful material within minutes of it appearing. Yet many of these cases may never reach a courtroom, and according to one trust and safety specialist, the reason has little to do with whether harm occurred.
“The barrier is rarely whether the harm was real,” says Shigemasa Sasaki, founder of RightsTech Inc. and a specialist in digital evidence infrastructure. “It is whether the proof still exists in a form the law can actually use.”
That distinction, between identifying harm and proving it, sits at the center of a problem the industry has been slow to address. Detection technology has advanced rapidly. The infrastructure needed to turn a detection into court-admissible evidence has not.
The evidence gap that few people see
Sasaki’s perspective is grounded in data rather than anecdote.
The dataset comes from Japan-based legal operations, Sasaki notes, and should be viewed as directional rather than definitive.
“Identification creates a false sense of progress,” he says. “People find the harmful post, capture a screenshot, and assume the hard part is over. In legal terms, it is just beginning.”
What the courts actually require
At the heart of the issue, Sasaki points to the authentication standard. Under the Federal Rules of Evidence, the party offering an item must produce sufficient proof to support a finding that it is what they claim it to be. For online material, he explains, that abstract requirement can be understood through several practical considerations: a verified timestamp of capture, a documented point of origin such as the source URL and the account or system that produced it, metadata that anchors the file in time and place, and, increasingly, a cryptographic hash confirming the material has not changed since it was preserved.
“Tamper-resistant is not a marketing term,” Sasaki says. “It means write-once storage and a hash recorded at the moment of capture. That is what allows a court to confirm that the file presented as evidence is identical to the one originally collected.”
The potential consequences of skipping that step are becoming increasingly apparent. As synthetic media becomes more convincing, courts may be applying greater scrutiny about what they will accept as authentic, which raises, rather than lowers, the bar for how evidence is captured and preserved.
Detection is not proof
Much of the trust and safety industry, in Sasaki’s view, has optimized for the wrong question. Tools that detect deepfakes and flag online abuse answer one question well: does harm exist? Legal systems ask a different question: can it be proven?
Many products stop at the first question. They confirm harm without preserving or structuring the underlying evidence, leaving a gap between discovery and enforcement that no amount of detection accuracy can close.
Addressing that gap may require three capabilities working together at the moment of capture: detection, meaning accurate identification of harmful content; preservation, including verified timestamps and tamper-resistant storage of the original material; and legal structuring, so evidence reaches a professional in a form they can act on without rebuilding it from scratch.
“If any one of those fails, the case may never proceed,” Sasaki says. “And because they all depend on the very first step, you usually cannot fix it after the fact.”
Designing for the handoff
The platforms most likely to improve outcomes, Sasaki believes, treat detection, preservation, and legal preparation as a single continuous workflow rather than three disconnected stages. When those steps exist within one process, evidence carries a consistent record from capture to courtroom, addressing some of the issues opposing counsel may raise when questioning authenticity.
It is the principle behind beME, the Japan-based service Sasaki developed to put that idea into practice. The platform combines harmful-content detection with timestamped, tamper-resistant preservation and a referral pathway to legal professionals. Former investigators help ensure documentation remains procedurally sound from the outset.
Two operational choices, he says, reflect the way harm behaves in practice: continuous monitoring, because exposed material can reappear or spread across platforms at any hour, and rapid first response, because acting within the first hours makes it possible to preserve the record while it still exists and limit further distribution.
Building the standard by example
Sasaki’s emphasis on verifiable process reflects a career built in fields where documentation and compliance determine outcomes. Over more than 25 years in life sciences and healthcare, he has been involved in multiple mergers and acquisitions, including the $85 million acquisition of OriCiro Genomics by Moderna.
He later founded RightsTech and developed beME, applying the same principles, provenance, integrity, and a documented record to digital evidence rather than biological material.
That path, from regulated science to trust and safety, informs a conviction he returns to often: the credibility of a case is engineered long before it is ever filed, through the design decisions made when evidence is first captured.
The longer arc
The legal system has navigated similar transitions before, and Sasaki views the current moment as part of a familiar pattern. Effective December 2017, the Federal Rules of Evidence were amended through Rules 902(13) and 902(14) to allow certain electronic records and forensic copies to be self-authenticated through written certification rather than live testimony, with matching hash values serving as proof that a copy is identical to its original.
The change reflected growing recognition that the traditional witness-by-witness approach may not always align with the scale and complexity of digitally created and stored information.
Earlier waves of technically complex litigation encountered similar challenges, as methods for managing evidence continued to develop alongside the underlying claims. The shift now underway around deepfake and online-abuse evidence, Sasaki argues, is the next phase of that process. The industry’s task is to treat evidence infrastructure as a core component of trust and safety rather than an afterthought.
“The harm is no longer the hard part to establish. The proof is,” Sasaki says. “Addressing that gap may influence whether these cases can proceed effectively.”
The information provided in this article is for general informational and educational purposes only. It is not intended as legal advice. Readers should not rely solely on the content of this article and are encouraged to seek professional advice tailored to their specific circumstances. We disclaim any liability for any loss or damage arising directly or indirectly from the use of, or reliance on, the information presented.
VentureBeat newsroom and editorial staff were not involved in the creation of this content.