For years, Bitcoin’s greatest strength has also been its simplest promise: once a transaction is written to the blockchain, it stays there. No bank can reverse it, no government needs to approve it, and no financial institution sits between buyer and seller deciding whether the transaction should proceed. The network verifies ownership cryptographically and, once consensus is reached, the transaction becomes part of Bitcoin’s permanent history. It is an extraordinarily powerful idea. But what happens when the person making the transaction is a thief? Recent events surrounding Coldcard, one of the better-known Bitcoin hardware wallets, provide an uncomfortable answer. Bitcoin’s blockchain does not know that a bitcoin has been stolen. It only knows that it has been transferred. Therein may lie Bitcoin’s Achilles heel.

Bitcoin has endured another difficult period in 2026. After the extraordinary institutionalisation of cryptocurrency that accompanied the arrival of spot Bitcoin ETFs and increasing involvement from mainstream financial institutions, the assumption that Bitcoin had finally matured into an established financial asset appeared increasingly plausible. Yet Bitcoin has fallen sharply during 2026. The decline cannot reasonably be attributed to one event. Changing expectations surrounding interest rates, institutional positioning, leveraged trading, geopolitical uncertainty and broader changes in investors’ appetite for risk have all played a role. Nevertheless, another development occurring alongside the decline deserves considerably more attention. Bitcoin has once again demonstrated just how difficult it is to secure an asset whose defining characteristic is that ownership ultimately depends upon possession of cryptographic credentials.

The Coldcard incident illustrates the problem particularly well. Hardware wallets exist because leaving cryptocurrency on an exchange creates an obvious contradiction. Bitcoin was designed to remove the need for financial intermediaries, yet leaving Bitcoin with an exchange simply replaces the bank with another institution. The alternative is self-custody, in which the investor controls the private keys required to move the Bitcoin. Hardware wallets such as Coldcard are specifically designed to keep those credentials away from internet-connected computers and therefore beyond the reach of conventional hackers. That, at least, is the theory. In late July, reports emerged of a serious exploit involving older Coldcard devices. Initial reports suggested that approximately 594 Bitcoin, worth around $38 million at the time, had been stolen, while subsequent analysis suggested the losses were substantially larger, potentially reaching around 1,367 Bitcoin, or approximately $89 million.

There is an important qualification here. Bitcoin itself was not hacked. The Bitcoin blockchain continued operating. That distinction will undoubtedly be emphasised by Bitcoin’s defenders and, technically, they are correct. Yet this creates a much more interesting problem because the blockchain worked. Imagine somebody steals the credentials required to access a conventional bank account and attempts to empty it. Banks have developed enormous institutional infrastructures for dealing with precisely this situation. Transactions can be flagged, accounts can be frozen, payments can sometimes be reversed, police can become involved, courts can determine ownership and financial institutions can be ordered to return assets. None of these systems is perfect. Fraud remains enormous, banks make mistakes and innocent customers sometimes struggle for months to recover stolen money. Bitcoin’s innovation was partly to remove the need for these intermediaries by demonstrating ownership cryptographically.

Cryptography, however, answers a surprisingly narrow question: does this person possess the credentials necessary to authorise this transaction? It cannot answer another question that human societies have spent thousands of years developing institutions to resolve: should this person possess them? If somebody steals a private key, Bitcoin cannot distinguish the thief from the owner. The thief presents the correct cryptographic credentials, the network verifies them, the Bitcoin moves and the blockchain permanently records what happened. Everything works exactly as designed. This is the paradox at the centre of Bitcoin. Blockchain does not necessarily fail when Bitcoin is stolen; rather, it can succeed indiscriminately.

Blockchain is frequently described as extraordinarily secure because historical transactions are exceptionally difficult to alter. That remains true, but immutability has an uncomfortable corollary. A fraudulent transaction that successfully satisfies the rules of the network can become exceptionally difficult to reverse as well. The blockchain does not understand theft, coercion or fraud. It does not know whether someone was threatened into revealing a password, whether malware extracted a private key or whether a hardware vulnerability allowed an attacker to obtain information that should have remained secret. It understands cryptographic validity. Bitcoin therefore solves one problem brilliantly while potentially exposing another. It makes the ledger extraordinarily difficult to falsify, but it does not necessarily make ownership extraordinarily difficult to steal. Once ownership has been stolen, blockchain’s celebrated immutability can become an advantage to the thief. The same infrastructure designed to prevent a bank, government or malicious intermediary from reversing a legitimate transaction also makes it difficult for those institutions to reverse an illegitimate one.

There is a temptation here to make an overly simplistic argument that Bitcoin causes crime. It does not. Cash facilitates crime, banks facilitate crime, companies facilitate crime and gold has been stolen, smuggled and used to conceal wealth for centuries. Technology rarely creates the underlying human motivations behind criminal behaviour. Bitcoin does, however, possess characteristics that can be attractive where individuals wish to move value without conventional intermediaries. It is global, permissionless and difficult to reverse; enormous amounts of value can be controlled without possession of a corresponding physical asset, while addresses do not inherently reveal the real-world identity of their owners. There is an important counterargument: Bitcoin’s blockchain is public. Transactions remain visible, creating a permanent forensic trail that has allowed law-enforcement agencies and blockchain-analysis companies to follow cryptocurrency across the network. Bitcoin therefore is not simply anonymous money. Nevertheless, the blockchain being able to tell us where stolen property moved is very different from being able to get that property back.

This exposes another contradiction in the proposition that Bitcoin is “trustless”. Ordinary Bitcoin owners must actually trust an extraordinary number of things. They must trust the software generating their keys, the hardware wallet storing them, its firmware, its supply chain and the computer interacting with it. They must trust that malware has not compromised another component of the system. They must safely preserve seed phrases and, ultimately, trust themselves not to make a catastrophic mistake. Bitcoin therefore has not eliminated trust so much as relocated it. Instead of trusting a bank to maintain the ledger and protect the account, the owner assumes responsibility for an increasingly complex technological chain. For technically sophisticated users that may be an acceptable exchange. For hundreds of millions of ordinary consumers, it is another question entirely. The more valuable Bitcoin becomes, the greater the economic incentive to attack every component surrounding the supposedly impregnable blockchain. Hackers do not necessarily need to break Bitcoin; they merely need to break the humans and technologies that hold the keys.

There is, however, an important problem with blaming Bitcoin’s recent decline entirely on these vulnerabilities: Bitcoin is not the only asset investors have been selling. Gold has also experienced significant selling pressure at various points during 2026. Equities and other risk assets have experienced bouts of volatility, while investors have been repositioning across bonds, commodities and cash-like instruments. This presents an important counterfactual. If Bitcoin were collapsing while gold surged relentlessly, it would be tempting to conclude that investors were specifically abandoning digital scarcity for the security of physical scarcity. The actual picture is considerably messier, suggesting that at least part of Bitcoin’s decline may have relatively little to do with Bitcoin itself. The relevant question therefore changes from simply asking why investors are selling Bitcoin to asking what investors are buying instead.

There is evidence of a remarkable movement towards liquidity. US money-market fund assets have climbed to extraordinary levels, with Investment Company Institute figures released in early August showing assets increasing by more than $55 billion in a single week, taking the total to around $7.9 trillion. Longer-term Federal Reserve data similarly show money-market assets running substantially above their levels only a year earlier. Recent international fund-flow data also indicate significant movements into money-market funds alongside flows into bonds, selected equities and precious-metal funds. This does not allow us to claim that a dollar withdrawn from Bitcoin has subsequently appeared in a particular money-market fund; financial flows cannot be traced that neatly from aggregate data. It does, however, provide evidence of a broader preference for liquidity occurring at the same time as investors have reduced exposure to some volatile assets.

Perhaps investors are therefore not simply buying cash. They may be buying optionality. Cash is normally considered an unproductive asset because when markets are rising rapidly, holding it carries an obvious opportunity cost. Cash becomes considerably more interesting when uncertainty rises, particularly when cash-like investments themselves provide attractive yields. An investor holding money in a money-market fund does not need to decide today whether Bitcoin, gold, equities or bonds represent the next great opportunity. They can wait. If Bitcoin falls further, they can buy Bitcoin later. If equities collapse, they can buy shares. If gold retreats, they can buy gold. If economic conditions stabilise, they can redeploy capital. Liquidity therefore possesses an option value: the investor has effectively purchased the ability to make tomorrow’s investment decision using tomorrow’s information.

This also means we should be cautious about interpreting simultaneous declines in Bitcoin, gold or other assets. It is conceivable that a severe Bitcoin sell-off can generate selling elsewhere. Leveraged investors facing cryptocurrency losses may need to liquidate profitable positions in gold, equities or other assets to meet margin requirements or restore portfolio liquidity. This phenomenon is familiar from previous financial crises, when investors sometimes sell the assets they can sell rather than those they necessarily want to sell. That can produce the apparently irrational spectacle of safe-haven assets declining during periods when investors are supposedly seeking safety. There is not presently enough evidence, however, to conclude that Bitcoin selling itself caused recent movements in gold. The more defensible interpretation is that both markets are being influenced by a broader reassessment of liquidity, risk, interest rates and portfolio positioning.

This qualification makes the Bitcoin story more interesting rather than less. Bitcoin is now sufficiently integrated into conventional financial markets that distinguishing a “crypto crash” from an ordinary financial-market correction is becoming increasingly difficult. Bitcoin wanted to escape the financial system, yet increasingly the financial system has absorbed Bitcoin. The growth of ETFs is perhaps the clearest example. One logical response to hardware-wallet vulnerabilities is simply to abandon self-custody. An investor might reasonably conclude that safeguarding private keys, firmware, seed phrases and physical hardware is too complicated, and instead purchase Bitcoin through an ETF. A regulated financial institution holds the underlying assets, professional custodians manage security and investors hold shares through conventional brokerage accounts.

There is considerable irony in this evolution. Bitcoin began with the proposition that we did not need banks. Then came exchanges, specialist custodians, institutional storage and regulated investment products. Eventually, some of the largest financial institutions in the world began offering investors exposure to Bitcoin. The technology designed to eliminate trusted intermediaries has gradually reconstructed an ecosystem filled with trusted intermediaries. That does not necessarily mean Bitcoin has failed. Quite the opposite: it may demonstrate Bitcoin’s extraordinary ability to survive and adapt. Yet its survival may require compromising some of the philosophical principles upon which Bitcoin was founded.

The Coldcard incident will not destroy Bitcoin, nor does it demonstrate that Bitcoin’s blockchain has been broken. The deeper lesson is considerably more uncomfortable. Technology can replace some of the functions performed by institutions, but it cannot necessarily replace the reasons those institutions exist. Banks are inefficient, regulators can be bureaucratic, governments can abuse financial power, and financial intermediaries charge fees, make mistakes and sometimes fail spectacularly. Bitcoin offered an elegant technological response by replacing institutional trust with mathematical verification. Yet human economies require something mathematics alone cannot provide: judgement. Was this transaction authorised freely? Was this property stolen? Was somebody deceived? Who legitimately owns the asset? Should a transaction be reversed? Who carries the loss when something goes wrong? These are not cryptographic questions. They are institutional ones.

That may ultimately be blockchain’s Achilles heel. Its greatest achievement is creating a ledger extraordinarily resistant to human interference. Its greatest weakness may be creating a ledger that is extraordinarily resistant to human intervention even when intervention is precisely what is required. The current Bitcoin sell-off cannot therefore be explained simply by Coldcard, hacking or fears surrounding cryptocurrency. The enormous accumulation of money in cash-like instruments suggests something broader is occurring as investors reassess risk and place an increasingly high value on liquidity. Bitcoin may recover, as it has after far more dramatic collapses. Gold may rise again, and capital currently sitting in money-market funds may eventually flood back into risk assets. But Coldcard leaves behind a question that will survive regardless of where Bitcoin trades next. Bitcoin was designed to create money that did not require us to trust institutions. Nearly two decades later, its evolution may be demonstrating something entirely different: perhaps institutions were never merely an inefficient obstacle standing between people and their money. Perhaps some of them were there for a reason. And Bitcoin’s greatest technological strength may ultimately be what forces us to rediscover why.

Table 1. Crypto and their price falls.

Cryptocurrency	8 Aug 2025	8 Aug 2026*	1-year change
BNB	$793.44	~$592.58	−25.3%
Bitcoin (BTC)	$116,688.73	~$64,945	−44.3%
Ethereum (ETH)	$4,009.85	~$1,918	−52.2%
Solana (SOL)	$176.76	~$74.67	−57.8%
XRP	$3.287	~$1.020	−69.0%
Dogecoin (DOGE)	$0.2302	~$0.0698	−69.7%
Avalanche (AVAX)	$23.80	~$6.47	−72.8%
Cardano (ADA)	$0.791	~$0.198	−75.0%

*8 August 2026 prices are intraday/current rather than final daily closes, so the percentages will move slightly before the day ends. Historical closes are from Yahoo Finance; for example, Bitcoin closed at $116,688.73 on 8 August 2025, Ethereum at $4,009.85, XRP at $3.2871 and Solana at $176.76. BNB, Dogecoin, Cardano and Avalanche similarly closed at $793.44, $0.23021, $0.79118 and $23.7974 respectively.

Fig. 1. Major crypto losses.

Every age has its gatekeepers.

Sometimes they have worn robes. Sometimes they have worn powdered wigs. Sometimes they have worn academic gowns, editorial titles, or corporate logos. Whatever their appearance, they have often shared one characteristic: they controlled access to knowledge.

Today, artificial intelligence has disrupted that arrangement more profoundly than perhaps any technology since the invention of the printing press. Yet the resistance to AI often follows a familiar pattern. It is presented as concern over accuracy, reliability, ethics or quality. These concerns deserve serious discussion. But I wonder whether they are also masking something much older: a discomfort with the democratisation of knowledge itself.

To explore that possibility, we first need to ask a deceptively simple question.

What is knowledge?

The Greeks wrestled with this question long before universities, peer review or digital libraries existed. Plato famously distinguished between belief and knowledge, suggesting that genuine knowledge required more than simply possessing information. His Allegory of the Cave remains one of the most powerful descriptions of humanity’s struggle to move from shadows towards reality.

Aristotle shifted the discussion towards observation and experience, arguing that knowledge grows through careful examination of the world. Other Greek philosophers questioned whether certainty was ever fully attainable, reminding us that knowledge has always been something to pursue rather than something we permanently possess.

What is striking is that none of these debates assumed knowledge should belong only to an elite. The debate concerned how we know, not who was permitted to know.

For most of human history, however, access to knowledge became inseparable from access to power.

Books were scarce. Literacy was limited. Universities admitted only a tiny fraction of society. Libraries were often private collections. Education depended upon wealth, geography and social class. Learning frequently required finding a teacher, joining an institution or entering a profession. Knowledge was not simply discovered; it was accessed through permission. Foucault argued that power and knowledge cannot be separated. Knowledge does not simply emerge because it is true; it is recognised through institutions that decide which voices are heard, which evidence is accepted and which ideas become legitimate (Foucault, 1977; 1980). If Foucault is correct, then the question is not simply whether AI generates accurate knowledge. The more interesting question is whether AI disrupts the traditional institutions that have historically determined who is permitted to participate in the production, communication and legitimisation of knowledge.

Colonialism amplified this asymmetry.

Colonial power was never exercised solely through military or economic dominance. It also depended upon controlling maps, scientific knowledge, language, education and historical narratives. Those who defined knowledge frequently defined civilisation itself. Entire cultures found their histories rewritten, their expertise dismissed and their own systems of knowing regarded as inferior.

Knowledge became another resource to be extracted, controlled and monetised. And although formal colonial empires have largely disappeared, many of their intellectual structures remain recognisable today.

Access to scientific journals often depends upon paywalls. Editorial boards determine which voices become part of the academic record. Professional language can become a barrier as much as a means of communication. Expertise is essential, but expertise can also become institutionalised in ways that unintentionally exclude those without the financial or social capital to participate.

None of this necessarily reflects malicious intent. It is simply how knowledge institutions evolved.

Then the internet arrived.

Dictionaries became searchable within seconds. Encyclopaedias no longer occupied shelves but entire servers. Wikipedia emerged with an almost absurd proposition: perhaps millions of people could collaboratively build the world’s largest encyclopaedia.

Its critics were immediate and often dismissive.

“It can be edited by anyone.”

“It cannot be trusted.”

“It is too easy.”

These criticisms contained elements of truth. Wikipedia has made mistakes, suffered vandalism and required continual moderation. Yet many critics overlook the extraordinary sophistication that now exists behind the platform. Articles on politically sensitive or scientifically important topics are often protected, monitored by experienced editors, supported by extensive referencing requirements and subject to continuous review. In many cases, Wikipedia’s greatest strength is not that anyone can edit it, but that everyone can see the discussion surrounding those edits.

The criticism, then, cannot simply be that knowledge has become easier to obtain, but rather artificial intelligence may represent the next stage in this evolution.

Unlike a search engine, AI does not merely retrieve information. It helps people navigate it. It explains difficult concepts, translates specialist language into everyday English, compares competing viewpoints and adapts explanations to the learner rather than forcing the learner to adapt to the material.

That changes something fundamental.

For centuries, one of the greatest educational advantages available to privileged individuals was access to patient human tutors. Someone who could explain difficult ideas differently each time until understanding emerged. AI now provides a version of that experience at almost no marginal cost.

This is where I believe many current debates become philosophically interesting.

When critics argue that AI makes writing “too easy”, what exactly is becoming easier? Is it the thinking? Or is it simply the expression of thought?

These are very different things.

Having an original idea has never depended upon flawless grammar. Creativity has never belonged exclusively to those fortunate enough to receive elite educations or decades of editorial guidance. Many people possess remarkable insights but lack confidence in academic writing, formal English or professional presentation.

AI changes that balance.

It allows individuals to communicate ideas that may previously have remained trapped behind linguistic, educational or economic barriers. The technology does not generate originality by itself. Rather, it allows originality to become visible.

Seen this way, AI represents less a replacement for human intelligence than a redistribution of intellectual opportunity.

Of course, AI should not be treated uncritically. It hallucinates. It reflects biases in training data. It sometimes presents uncertainty with unwarranted confidence. These limitations require informed users and continued human judgement.

But none of those limitations justify resisting wider access to knowledge itself.

Perhaps the real question is not whether AI threatens expertise.

Perhaps it threatens monopoly.

Every technological revolution that has expanded access to knowledge has been criticised in similar ways. The printing press threatened scribes. Cheap paperbacks threatened traditional publishing. Public libraries threatened commercial lending libraries. Wikipedia threatened printed encyclopaedias.

Now AI threatens something even larger: the assumption that understanding should remain difficult primarily because access has been difficult in the past. If knowledge genuinely exists to improve humanity, then widening access should be celebrated rather than feared.

The democratisation of knowledge has never been comfortable for those who benefited from its scarcity.

Artificial intelligence may simply be the latest chapter in a story that began long before computers existed—a story not about machines replacing people, but about removing the gates that have too often stood between people and knowledge.

References:

Foucault, M. (1977) Discipline and Punish: The Birth of the Prison. Translated by A. Sheridan. London: Allen Lane.

Foucault, M. (1980) Power/Knowledge: Selected Interviews and Other Writings 1972–1977. Edited by C. Gordon. Brighton: Harvester Press.