The ledger does not lie, only the interpreters do.
SoftBank Group announced the appointment of Yossi Cohen, former director of Mossad, as a strategic advisor for its artificial intelligence investment strategy. The press release, sparse on details, frames the move as a response to the growing complexity of geopolitical risks in AI deployment. But a deeper reading of the balance sheet—and the human capital behind it—reveals a structural shift in how capital allocators are now valuing the intersection of intelligence, security, and technology.
This is not a personnel change. It is a signal that the next phase of AI investment will be defined by adversarial thinking, not just model performance.
Context: The Global Liquidity Map and SoftBank’s Pivot
SoftBank’s transformation from a diversified telecom and internet investment conglomerate into a pure AI play began in earnest after the Vision Fund’s post-WeWork restructuring. Founder Masayoshi Son has repeatedly stated that artificial general intelligence (AGI) will arrive within a decade, and that SoftBank must become the infrastructure provider for that era. The key pillar is Arm Holdings, the chip architecture company that SoftBank acquired in 2016 and still controls a majority of after its 2023 IPO. Arm’s instruction set powers nearly every mobile AI chip and an increasing share of server-grade AI accelerators. This gives SoftBank a unique, non-dilutable position in the AI supply chain: it does not just invest in AI companies; it owns the architectural layer that many of them depend on.
However, the capital allocation strategy has evolved. The Vision Fund’s early bets on ride-sharing, co-working, and food delivery proved volatile. The lesson was clear: without a controlling stake or a defensible moat, even large investments can be eroded by market sentiment. The AI pivot, therefore, is not just about sector selection—it is about structural control. SoftBank now seeks to build an integrated ecosystem where Arm, data centers, and AI applications form a closed loop. Cohen’s appointment is the missing piece: a mechanism to assess and manage the asymmetric risks that come with operating at the edge of national security.
Core: The Intelligence-Adjacent Investment Thesis
From my own experience auditing over 50 ICO projects during the 2017 mania, I learned that the most dangerous blind spots are not technical but behavioral. A smart contract can be formally verified, but the intentions of the team behind it cannot. The same principle applies at the institutional scale. SoftBank’s challenge in AI is not finding good technology—it is distinguishing between technologies that are genuinely transformative and those that will create catastrophic tail risks. Cohen’s background is not in AI engineering; it is in evaluating threats from state-level actors, running covert operations, and understanding the gap between public claims and private capabilities. That skill set is directly applicable to the current AI landscape, where companies routinely overstate the capabilities of their models and understate the risks of misuse.
Consider the three layers where Cohen’s intelligence experience adds value:
- Due Diligence on AI Security Startups – The field of AI safety, red-teaming, and model alignment is becoming a multi-billion dollar sector. But many of these startups are founded by researchers with academic backgrounds, not by people who have operated against sophisticated adversaries. A former intelligence chief can assess whether a company’s security claims are credible or merely marketing. Based on my 2020 liquidity stress tests on DeFi protocols, I know that the most dangerous vulnerabilities are often the ones that founders themselves do not see. Cohen’s network can provide independent verification of a startup’s defensibility.
- Geopolitical Risk Hedging – AI infrastructure is inherently geopolitical. The supply chain for GPUs, the location of data centers, and the regulatory regimes for training data all intersect with national security interests. SoftBank operates in multiple jurisdictions, including China, where it has invested in companies like ByteDance and Didi. A former Mossad advisor can help map the implicit risks of operating in certain markets—not just legal risks, but the unspoken expectations of intelligence agencies. This is not about paranoia; it is about capital preservation. As I wrote during the 2022 bear market, rebalancing is not panic; it is preservation.
- Access to the Israeli AI Security Ecosystem – Israel produces a disproportionate number of world-class cybersecurity and AI defense startups, many of which are founded by alumni of Unit 8200 and other intelligence units. Cohen’s personal network gives SoftBank a privileged access point that no other Western fund can replicate. This is a classic example of what I call historical liquidity mapping: the ability to tap into a flow of deals that is not visible on public radars. The value of this access cannot be overstated, especially in a market where the best deals are often closed before they reach a competitive process.
Contrarian: The Decoupling Thesis and the Risk of Over-Indexing
The conventional narrative is that Cohen’s appointment makes SoftBank a stronger, more informed investor. The contrarian view is that it introduces a new category of risk that could decouple SoftBank from the broader AI ecosystem.
First, there is the question of trust. Many AI researchers and developers are deeply skeptical of any entanglement with intelligence agencies. The open-source community, which drives much of the innovation in AI, could view SoftBank as contaminated. This could affect the willingness of top AI talent to work with SoftBank-backed companies. I recall a similar dynamic in the early days of crypto, when projects that accepted funding from questionable sources had difficulty recruiting developers. Code is law, but humans are the bug.
Second, the geopolitical backlash could be material. SoftBank’s largest limited partner is the Public Investment Fund of Saudi Arabia. Saudi Arabia and Israel do not have formal diplomatic relations, and while the Abraham Accords have improved ties, the appointment of a former Mossad chief could strain the relationship. If PIF reduces its commitment to future Vision Funds, SoftBank’s capital base could shrink. Liquidity dries up when trust evaporates.
Third, there is the risk of over-optimization. By focusing on security and intelligence, SoftBank may miss the next wave of AI innovation that comes from open, decentralized, or non-Western sources. The contrarian bet is that the most valuable AI applications will emerge from environments that are not security-centric—just as the most valuable crypto protocols emerged from open-source communities, not from walled gardens. SoftBank’s pivot to intelligence could make it the most sophisticated investor in AI security, but it could also make it blind to the next paradigm shift.
Takeaway: Positioning for the Intelligence Cycle
Every bull run is a tax on due diligence. The AI investment cycle is no different. SoftBank’s move to hire Yossi Cohen is a recognition that the next phase of AI will be defined by asymmetric risks, not by incremental improvements in benchmark scores. The question is not whether this makes SoftBank more informed—it clearly does. The question is whether the added intelligence layer will create a feedback loop that distorts its investment thesis, pulling it toward defense and security at the expense of the broader, more creative currents of AI development.
For the crypto native, the parallels are instructive. We have seen what happens when centralized capital allocators fail to account for network effects and community trust. The decentralized world thrives on permissionless innovation, but it also requires a different kind of risk assessment. SoftBank’s intelligence gambit is a reminder that even in a world of smart contracts and on-chain governance, the most important signals are still the ones that are not written in code.