Is Bitcoin Safe? What Is Protected and What Is Not

Splitting "safe" into four layers

• The question "is Bitcoin safe?" cannot be answered as asked, because the answer flips depending on which safety is meant. The ledger is extremely resistant to tampering, and at the same time there is no mechanism anywhere that helps someone who sent to the wrong address. The same word points at entirely different layers.
• This page separates the question into four. Argue without separating them and two facts look contradictory — "not broken in 17 years" and "large losses every year" — when in fact both are true.

• Each of these four layers has its own page on this site. Protocol weaknesses are covered in "Bitcoin Vulnerabilities", the custody layer in "Wallets and Security", the counterparty layer in "Bitcoin Scams and Self-Defense", and institutional failure in "Bitcoin Incidents and Turning Points". This page is the entrance: its purpose is to let you tell which layer a given claim is about.
• The one-line summary is this. The protocol layer has not been broken in more than 17 years, and most of the bitcoin ever lost was lost in the other three.

The protocol layer — what 17-plus years has held

• Since the network started in 2009, the ledger itself has never been tampered with. Two mechanisms do most of the work: Proof of Work, which demands real computation and electricity to extend the chain, and independent validation, in which every full node checks each block against the rules for itself.
• The boundary of what an attacker can do is specific. An attacker holding a majority of hash power can reverse their own transactions (double-spend), keep particular transactions out of blocks (censorship), and orphan other miners' blocks. What they cannot do is move coins without the private key, create BTC that does not exist, or unilaterally change the 21 million cap — full nodes reject rule-breaking blocks even when the majority produced them.
• No 51% attack has occurred on Bitcoin. The pool GHash.io briefly exceeded 50% in 2014 and prompted debate, but no attack followed. As a rough measure of the cost, mempool.space put total network hash rate at roughly 913 EH/s implied by difficulty on 16 August 2026, and roughly 914 EH/s on a seven-day average. Hash rate cannot be observed directly; it is an estimate, and the shorter the averaging window the more it moves.
• "Has not been broken" is not the same as "cannot be broken". Whitepaper §11 calculates that an attacker with only 10% of hash power still catches up about 20% of the time at one confirmation; the same table falls to about 0.024% at six. The convention of treating six confirmations as final comes from that curve. Choosing a confirmation count is risk management against amount and counterparty, not a declaration of absolute safety.
• The live concern is mining pool concentration. On mempool.space's trailing-week tally for 16 August 2026 (1,005 blocks), the largest pool mined about 23% of blocks and the top three about 61%. Pools do not own the hash rate and participants can leave at any time, but the concentration itself is an observed fact. "Bitcoin Vulnerabilities" assesses it in more depth.

The exceptions — bugs that did happen, and why the network recovered

• "The protocol has not been broken" means, precisely, that the consensus mechanism has never been broken permanently. Cases where an implementation bug temporarily broke the rules are part of the public record.
• On 15 August 2010, block 74638 was found to contain a transaction creating over 184 billion BTC. The validation code had not anticipated the case where output values sum to an integer overflow. For a moment, the 21 million cap — the foundation of the whole design — was broken by an implementation defect. A patched client shipped roughly five hours after discovery, and the correct chain overtook the invalid one at block 74691.
• In September 2018, CVE-2018-17144 was fixed. A check for whether a single transaction spends the same output twice had been skipped; according to the official disclosure, under some conditions this also amounted to an inflation vulnerability that would let a miner create BTC beyond the cap. It was reported privately on 17 September, fixed in version 0.16.3 on 18 September, and disclosed in full on 20 September. There is no evidence it was exploited on mainnet.
• What these two cases show is that consensus rules are implemented as code, so an implementation bug can break the rule itself. Equally part of the record is that both were detected and fixed within hours to days, and the network recovered. Bitcoin Core formalized its vulnerability disclosure policy in July 2024, setting publication timing by severity.
• The reading to take away is that 17 years is not a history without bugs; it is a history of bugs being found, published, and fixed. The individual cases and the open problems are covered in "Bitcoin Vulnerabilities".

The custody layer — where most of it has actually been lost

• First among the things not protected is the key itself. Only the holder of the private key can move bitcoin, and no party can restore that key — not developers, not miners, not exchanges, not states. Because there is no privileged administrator, there is structurally no path along which proving your identity gets your coins back.
• By volume, losses in this layer exceed the losses from headline incidents. A 2025 study by the Bitcoin company River estimated roughly 1.57 million BTC lost in the course of self-custody and roughly 1.51 million BTC lost at exchanges. Widen the definition and the number grows: an analysis Chainalysis published in June 2020 put coins unmoved for five years or more at about 3.7 million BTC, roughly a fifth of the supply issued at the time. All of these are estimates, and "unmoved for a long time" is not proof of "inaccessible". "Why Lost Bitcoin Cannot Be Recovered" explains how to read the range.
• What has to be protected is narrow. A wallet handles three layers — private key, public key, and address. The address is meant to be published, and the public key becomes public as soon as you spend. Only the private key and the seed phrase it derives from must never be known to anyone else. BIP-39 defines seed phrase lengths of 12, 15, 18, 21, or 24 words; 12 and 24 are the ones in common use.
• Technical protection also has a physical limit — the so-called "$5 wrench attack", in which the holder is identified and coerced. Jameson Lopp's public tally of reported cases records 352 incidents since December 2014, with 85 in 2025 the highest annual count on record (54 so far in 2026 through early August; both counts as of 16 August 2026). As the author notes, the list is not exhaustive, so the real figure should be assumed higher. This is why "do not advertise what you hold" keeps appearing among the recommended practices.
• Choosing a custody method does not remove risk; it selects which risk you take on. The specific mechanics — hot versus cold, HD wallets, multisig and PSBT — are covered in "Wallets and Security". This site does not recommend any particular product or manufacturer.

The counterparty layer — what the incident record shows

• Almost every large historical loss happened in this layer. The common thread is that Bitcoin's cryptography and consensus were not broken; the operations of whoever was holding the coins were.
• Mt. Gox announced the loss of roughly 850,000 BTC in February 2014. About 200,000 BTC were found in an old-format wallet the following month, which puts the substantive loss at roughly 650,000 BTC. Repayment to creditors has proceeded under Japan's civil rehabilitation framework and was still incomplete as of August 2026; after several extensions, the deadline stands at 31 October 2026.
• In Japan, Coincheck lost roughly 58 billion yen worth of NEM in 2018 and compensated about 260,000 affected users with roughly 46.3 billion yen from its own funds. On 31 May 2024, 4,502.9 BTC (roughly 48.2 billion yen at the time) left DMM Bitcoin; the company compensated customer assets in full and then wound the business down. On 21 February 2025, Bybit lost 401,347 ETH (roughly 1.4–1.5 billion dollars at the time), and the method was not key theft but rewriting what the signers saw on the approval screen — a case showing that keys are not the only thing that has to be protected.
• When the counterparty is an individual rather than a business, the failure mode is fraud. Japan's National Police Agency confirmed figures for 2025 record 15,168 cases of SNS-based investment fraud and SNS-based romance fraud combined, with losses of 183.43 billion yen — up 48.2% in cases and 44.2% in losses year on year. Crypto-assets are increasingly the form in which victims hand money over: within SNS-based investment fraud, crypto-transfer cases numbered 1,898 with 21.55 billion yen, and even in traditional telephone-based special fraud, crypto-transfer cases reached 1,243 with 19.69 billion yen. The National Consumer Affairs Center logged 8,132 crypto-asset consultations in fiscal 2025.
• None of these figures show that Bitcoin was broken. What they show is that however strong the cryptography, the path in which a person is deceived into sending voluntarily cannot be closed. The structure of the methods and how to spot them is covered in "Bitcoin Scams and Self-Defense"; the incidents themselves in "Bitcoin Incidents and Turning Points".

The institutional layer — the protection Japan offers, and its limits

• What follows describes the framework in force as of August 2026. The framework is in transition: moving crypto-asset regulation from the Payment Services Act to the Financial Instruments and Exchange Act is settled as law, but that part had not taken effect as of this writing.
• Exchanging crypto-assets for fiat currency as a business in Japan requires registration, and operating without it violates Article 63-2 of the Payment Services Act. The Financial Services Agency's register listed 26 providers as of 30 June 2026, and the list is updated as registrations and withdrawals occur. Penalties for unregistered operation were raised on 12 August 2026 to imprisonment of up to ten years, among other measures.
• One premise matters. The FSA states on the register itself that it neither guarantees nor endorses the value of these crypto-assets. Registration confirms that a minimum set of business rules is met; it is not an endorsement of any operator or any asset. A solicitation claiming to be "FSA-approved" can therefore be judged false on that point alone.
• On remedies after the fact, it is worth knowing exactly how far they go. Japan has several public helplines — the Consumer Hotline 188, the police consultation line #9110, and the FSA's Counseling Office for Financial Services Users at 0570-016811. The FSA office gives advice and refers you onward, but does not mediate, arbitrate, or conciliate. Proceed on the assumption that no public body will conduct refund negotiations on your behalf.
• Recovering funds sent as crypto-assets is extremely difficult. On-chain transfers cannot be reversed, and funds are usually dispersed through several exchange services within a short time. In other words, the institutional layer works for screening beforehand — checking whether an operator is registered — and barely at all for recovery afterwards. Secondary fraud offering to "get it back" for an advance fee is also widely documented, so the moment after a loss is exactly when a public helpline should come first.

Self-defense, in order of effect

• Given the sections above, the priority ordering falls out on its own. What works best is not the technically advanced measure but the dull habit that matches the layer where the most is being lost.

• The order has a reason. Items 1 and 2 address the paths through which, statistically, the most damage is being done right now. Item 3 takes the operation whose failure costs the most and has you experience it once while the stakes are small. Items 4 onward can be introduced in stages as holdings grow.
• None of this requires a specific product. This site recommends no particular wallet, exchange, or service; what is offered here is categories and procedures. Setup detail lives in "Wallets and Security", and the step-by-step checks for a transfer in "Sending Bitcoin".

Frequently asked questions

• "Has Bitcoin ever been hacked?" — The protocol itself has never been broken permanently. What gets reported as "a Bitcoin hack" is almost always a breach of an exchange or a surrounding tool. That said, as described above, implementation bugs have temporarily broken the rules. The two need to be kept apart.
• "Will quantum computers make it unsafe?" — What is affected is the elliptic-curve cryptography used for signatures; the effect on the hash function used in mining is considered limited. The coins at risk first are those whose public key is already exposed on-chain. Expert views on when this becomes a practical threat differ, and this site does not assert a date. "Bitcoin Vulnerabilities" covers it in detail.
• "Can a government freeze it?" — No party holds the power to freeze an on-chain balance itself. Exchange accounts and the on- and off-ramps to fiat currency, however, are regulated in every jurisdiction. The property of being hard to freeze applies to self-custody, and does not extend to assets you have deposited with someone. "Bitcoin Regulation" covers the wider picture.
• "If I send to the wrong address, can I get it back?" — As a rule, no. While the transaction is still unconfirmed it may be possible to replace it with one paying a higher fee, but there is no guarantee of success. Once confirmed, the only way back is for the recipient to send it back.
• "Which is safer, an exchange or self-custody?" — Neither is safer in general; they differ in which risk you take on. Deposit at an exchange and the risk of losing keys is replaced by the credit risk of that business; self-custody puts key management and inheritance planning on you. In River's estimate cited above, the amounts lost in each are roughly comparable.
• "Where should a beginner start?" — This site's answer is: read before you buy. Every topic here can be read end to end without buying any bitcoin, and "Start here" lays out a reading order.

In summary — how to answer "is it safe?"

• Here is what this page covered, split by layer into what can and cannot be established.

• Three structural points are worth holding onto. First, strength in the protocol layer does not compensate for weakness in the others. Second, most bitcoin ever lost was lost in the user-side layers, where what helps is procedure rather than technology. Third, remedies after the fact barely exist, so the value of any precaution is entirely in advance.
• The questions this site has no answer to are equally clear. Whether you should hold bitcoin, which wallet product or exchange is safe, and how much is a safe amount. These are value judgments or individual recommendations, and they sit outside educational commentary.
• This page is educational commentary, not investment advice and not individualized legal or tax advice. It is current as of August 2026; statistics, regulation, and network figures all change. Confirm against the primary sources cited here at the moment you actually need them.