The Mossad at the Table: SoftBank's Bet on Intelligence-Driven AI

Everyone is selling you the promise of AGI. No one is showing you the intelligence infrastructure required to secure it. Silence is the loudest audit. When SoftBank, the architect of the Vision Fund and the majority owner of Arm, announced the appointment of Yossi Cohen—former director of Mossad, Israel's national intelligence agency—as a strategic advisor for its AI investments, the market barely paused. The press release was a clinical note: Cohen will advise on AI strategy, portfolio security, and geopolitical risk. But the subtext is a revolution in how we think about AI capital. Trust the protocol, not the pitch. The pitch is that SoftBank is building for AGI. The protocol is that it is now weaving intelligence and sovereignty into the fabric of its investment thesis. Context matters. SoftBank is not a venture capital firm in the traditional sense. It is a holding company that owns Arm, the semiconductor architecture at the heart of nearly every AI chip. It manages the Vision Fund, a $100 billion vehicle that reshaped global tech funding. Its founder, Masayoshi Son, has publicly bet that AGI will arrive within a decade and that SoftBank will be its primary infrastructure provider. The shift from financial investor to industrialist is complete. But infrastructure requires more than silicon and capital. It requires security. The AI supply chain is a vector for state-level threats: model theft, data poisoning, adversarial attacks on deployed systems. SoftBank's introduction of a former intelligence chief suggests it has recognized that the next frontier of competition is not only performance but also resilience. Here is the core insight: the appointment is not a PR move. It is a structural bet on the convergence of AI and national security. Cohen's role is not to pitch startups; it is to evaluate them. In the world of AI, technical due diligence has become a black art. Most investors rely on academic credentials or benchmark scores. But those are outputs of a controlled environment. The real question is whether a model can survive deployment in a contested environment. That is a question that intelligence agencies have been asking for decades. Based on my experience auditing smart contracts during the DeFi summer, I learned that code alone does not guarantee trust. The same applies to AI. Code doesn't lie, but the people who write it do. A model’s security is not in its architecture but in its threat model. Cohen brings a threat model that is several orders of magnitude more sophisticated than the average VC analyst. But there is a stark ethical dimension. The same intelligence community that protects national secrets also operates in the shadows. Mossad has a history of covert operations, including cyber attacks and assassinations. Bringing that culture into the boardroom of a global investment firm creates a conflict of interest that no compliance manual can resolve. The crypto community understands this tension acutely. We have seen projects that claim to be decentralized while holding admin keys. We have seen exchanges that promise self-custody while lending user deposits. The pattern is the same: the pitch is freedom, the protocol is control. SoftBank's pitch is that it is building a safe AGI. But the protocol is that it is now integrated with the intelligence apparatus of a nation-state. That is not a bug; it is a feature. But it is a feature that the market must price in. Here is the contrarian angle: perhaps this is exactly what AI safety needs. The AI safety movement has been dominated by academic philosophers and engineers who have never faced a real adversary. They talk about alignment as if the only threat is a misaligned utility function. In reality, the most immediate threats are adversarial: malicious actors who will jailbreak, poison, or steal models. The people who understand adversarial thinking are intelligence officers. They have spent decades thinking about how to break systems and how to defend them. Cohen's appointment could be the most pragmatic move in AI safety today. It acknowledges that safety is not just a technical problem but a geopolitical one. By embedding intelligence expertise into its investment process, SoftBank may be able to filter out projects that are vulnerable to state-level attacks. That is a genuine value-add. But the risk of mission creep is high. Once intelligence agencies have a seat at the table, they tend to expand their influence. SoftBank's portfolio includes companies that process sensitive data—healthcare, finance, telecommunications. The line between security and surveillance is thin. We have seen this in the blockchain space: projects that start with transparency and privacy often end up with backdoors for regulators. The same pattern will repeat in AI. The introduction of a former Mossad chief does not guarantee that SoftBank will be more secure; it guarantees that it will be more connected to the national security apparatus. That is a trade-off that the industry must confront. I have been in this industry long enough to see cycles of hype and disillusionment. In 2017, I saw ICOs promise decentralization and deliver fraud. In 2020, I saw DeFi promise trustless finance and deliver collapsing yield farms. Now, I see AI promising intelligence and delivering opaque systems. The lesson is always the same: trust the protocol, not the pitch. The protocol here is that SoftBank is building a new kind of intelligence—not artificial, but organizational. It is combining capital, chip architecture, and spycraft. That is a powerful combination. But it is also a dangerous one. The takeaway is not a conclusion but a question. The question is: who audits the auditors? If SoftBank now has an intelligence chief advising on AI investments, who is advising the intelligence chief? The industry needs its own verification mechanisms. We need open-source audits of AI safety, not just from academics but from adversarial red teams. We need decentralized governance that prevents any single entity from controlling both the infrastructure and the intelligence behind it. The blockchain community has a role to play here. We have built systems that verify transactions without central authority. We can apply the same principles to AI. We can build proof-of-human-intent signatures that verify that an AI output is not being manipulated by a hidden agenda. We can build on-chain registries of model provenance. We can create a culture of transparency that makes it harder for intelligence agencies to co-opt AI under the guise of security. Silence is the loudest audit. The silence from the crypto community on this appointment is deafening. We are so focused on the next token launch that we ignore the structural shifts in the technology landscape. SoftBank is not just a fund; it is a prototype for the AI-industrial complex. If we do not build the tools to audit that complex, we will wake up to an intelligence infrastructure that is not accountable to anyone. The path forward is not to reject security but to insist that security is distributed. Trust the protocol, not the pitch. Build the protocol.