Late on Tuesday night, a passenger train travelling from Athens towards Thessaloniki collided head-on with a freight train near Tempi in central Greece. The passenger train was carrying around 350 people, many of them university students returning after a holiday weekend. Fifty-seven people were killed, making it the deadliest railway disaster in modern Greek history.
The immediate explanation appears straightforward. The two trains were travelling in opposite directions on the same track. A station master in Larissa has been arrested and has acknowledged responsibility for errors preceding the collision, while investigators are examining exactly how the trains were allowed to remain on a collision course for several minutes. If an employee incorrectly routes a train, individual responsibility obviously matters.
But the more information that emerges, the less satisfactory the phrase “human error” becomes as a complete explanation.
Complex systems are operated by humans. Humans eventually make mistakes. The purpose of a safety system is therefore not to construct a world in which nobody ever presses the wrong button. It is to make one person’s mistake insufficient to kill fifty-seven people.
That distinction is fundamental.
If an aircraft crashes whenever one pilot makes one incorrect decision, aviation safety has failed even if the pilot made a genuine error. Commercial aviation became extraordinarily safe partly because modern systems assume that pilots, controllers, mechanics and engineers are fallible. Procedures overlap. Computers generate warnings. Checklists create redundancy. Independent systems monitor one another.
The objective is not eliminating error.
It is preventing single-point failure.
Railways should operate through the same principle. A station master may route a train incorrectly. Signals should provide another layer of protection. Automated control systems can provide another. Communication between stations and drivers provides another. Organisational procedures provide another.
A catastrophe occurs when enough of those layers fail simultaneously.
That makes responsibility distributed.
The station master may still bear responsibility for the decision immediately preceding the crash.
The institution bears responsibility for constructing a system in which that decision could become catastrophic.
These are not competing explanations.
They exist at different levels.
Greece’s transport minister, Kostas Karamanlis, resigned the day after the collision. In his resignation statement, he said that after such a tragedy it was impossible for him to continue as though nothing had happened and that he considered accepting political responsibility necessary.
That is significant because political responsibility is different from criminal responsibility.
A minister does not need personally to have switched the railway points before accepting responsibility for the system underneath their ministry.
This distinction is increasingly absent from politics.
Politicians often behave as though resignation should occur only after personal wrongdoing has been proven. If they did not personally break the law, they argue, there is no reason to leave office.
But ministerial responsibility traditionally means something broader.
A cabinet minister receives political authority over an institution.
That authority brings responsibility for institutional performance.
If the only failures for which ministers are responsible are the actions they personally performed, political leadership becomes strangely asymmetrical: ministers receive credit for everything their departments achieve while accepting blame only for things they touched themselves.
Authority cannot reasonably work that way.
Karamanlis’s resignation does not establish that he caused the crash.
It recognises that leadership sometimes carries responsibility for failures whose causal chain extends far below the leader.
The harder question is where that chain should stop.
Greek railway workers had reportedly warned about safety problems before the collision. The network has faced years of difficulties involving ageing infrastructure, staffing and delayed modernisation. Greece had been attempting to install more advanced signalling and remote-control technology, but implementation had repeatedly been delayed.
Later technical assessments would conclude that systems such as the European Train Control System (ETCS) and the GSM-R railway communications network, if fully operational, could have provided additional safeguards against exactly this type of collision. Those systems were still not functioning fully even a year after the disaster.
That history changes the meaning of Tuesday’s accident.
A risk nobody anticipated is one category of failure.
A risk repeatedly identified but not corrected is another.
This is the difference between uncertainty and negligence.
Governments cannot prepare perfectly for every possible catastrophe. Public resources are finite, and policymakers have to decide which risks deserve investment.
But once a serious safety vulnerability is known, repeated postponement becomes a policy decision even when nobody formally votes to accept the risk.
Doing nothing is still an allocation choice.
Money was spent elsewhere.
Administrative attention went elsewhere.
Contracts were delayed.
Responsibility became dispersed.
Eventually the probability accumulated.
This is where institutional failures become difficult to see before catastrophe because no individual decision necessarily looks disastrous.
A signalling project is delayed six months.
Nothing happens.
A contract is renegotiated.
Nothing happens.
A maintenance position remains unfilled.
Nothing happens.
A temporary manual procedure continues for another year.
Nothing happens.
Each period without catastrophe appears to confirm that delay was tolerable.
Then two trains enter the same track.
This is a dangerous statistical intuition.
The fact that a system survived yesterday does not prove it was safe yesterday.
It may simply mean the low-probability failure did not occur.
Financial crises work similarly.
A bank can take excessive risk for years and appear exceptionally profitable until the risk materialises.
A building can violate seismic standards for thirty years and appear completely functional until the earthquake arrives.
A railway can operate with inadequate safeguards thousands of times before one mistake aligns with every missing layer of protection.
Success under normal conditions can therefore conceal fragility.
This is why safety regulation has to be forward-looking rather than outcome-based.
If we judge a system only after someone dies, we are always regulating one disaster late.
The protests now spreading across Greece reflect precisely this frustration. Demonstrators are not treating the collision simply as the mistake of one railway employee. They are connecting it to a broader perception that Greek institutions have tolerated deterioration in public infrastructure for too long. The crash would later become a lasting symbol of railway neglect and institutional failure, prompting some of the country’s largest demonstrations in years.
That anger has particular historical context because Greece spent much of the previous decade under extraordinary fiscal pressure.
After the sovereign-debt crisis beginning in 2009, Greece implemented severe austerity under agreements with European and international creditors. Public expenditure was constrained, wages and pensions were cut, unemployment rose dramatically and government institutions operated under sustained pressure.
It would be too simplistic to say “austerity caused the train crash.”
That would require a much more specific causal chain than is currently established.
But fiscal choices affect institutional capacity.
Infrastructure needs investment.
Safety systems need procurement.
Railways need staff.
Regulators need resources.
When governments reduce expenditure over long periods, the relevant question is not simply how much money was saved.
It is which forms of capacity disappeared with it.
This is the hidden side of fiscal consolidation.
Government expenditure appears in budgets as numbers.
State capacity often appears only when someone needs the service.
Cut ten inspectors and the budget improves immediately.
The cost of weaker inspection may remain invisible for years.
Reduce maintenance and the asset continues operating.
Until it does not.
This does not mean governments should never cut expenditure. Waste exists. Bureaucracies can become inefficient. Public debt creates genuine constraints, as Greece knows better than almost any European country.
The analytical problem is that not all government spending has the same social return.
Cutting an unnecessary administrative programme is different from postponing maintenance on critical infrastructure.
Aggregate spending numbers hide composition.
This is why debates about “big government” and “small government” are often much less useful than debates about capable government.
A state can spend enormous amounts inefficiently.
A state can spend relatively little while maintaining highly effective institutions.
The relevant question is whether public resources create reliable collective goods.
Railway safety is one.
This connects directly to the argument from Turkey two weeks ago. There, the question was how much of an earthquake disaster belongs to geology and how much to governance.
Greece removes geology from the equation almost entirely.
Railways are engineered systems.
Humans designed the tracks.
Humans scheduled the trains.
Humans wrote the procedures.
Humans purchased or failed to purchase safety equipment.
Humans staffed the stations.
When an engineered system produces catastrophe, describing the event as an “accident” can sometimes obscure how many institutional decisions preceded it.
That does not mean someone intended the deaths.
Intent and preventability are different concepts.
Many institutional disasters are nobody’s intention and everybody’s responsibility in small pieces.
That makes accountability politically difficult.
Courts are designed to identify individual legal liability.
Who violated which law?
Who acted negligently?
Who signed which document?
Those questions matter.
But systemic failure may not fit neatly inside criminal law.
Suppose twenty officials each contributed 5% to an institutional breakdown, and none individually committed an act severe enough to explain the outcome.
The cumulative system can still be catastrophically unsafe.
This is why inquiries, parliamentary oversight and administrative reform exist alongside criminal prosecution.
Different institutions answer different questions.
A criminal court asks who committed an offence.
An accident investigation asks how the system failed.
A parliament asks whether political leadership was adequate.
Voters ask whether they still trust the government.
Those judgements should not be collapsed into one another.
A minister can deserve political removal without deserving prison.
A railway employee can deserve criminal scrutiny without carrying the entire political responsibility.
An institution can require reform even if no single individual intentionally created the failure.
This distinction is essential because governments often respond to systemic disasters by finding one visible person to blame.
The station master becomes the explanation.
The contractor becomes the explanation.
The pilot becomes the explanation.
The police officer becomes the explanation.
Individual accountability can then function as a substitute for institutional accountability.
Punish the person at the end of the causal chain and the structure above them survives unchanged.
That is convenient.
It is not learning.
A serious investigation into Tempi therefore needs to ask not only what the station master did on Tuesday night, but why one station master’s actions possessed so much destructive potential.
Were automatic systems available?
If not, why?
Were staffing levels adequate?
Were workers trained appropriately?
What safety concerns had unions or employees raised?
Which modernisation projects were delayed?
Why were they delayed?
Who knew?
What happened after they knew?
Those questions move responsibility upwards without assuming in advance where it ends.
This is what systems thinking contributes to political accountability.
The same principle is used in aviation accident investigation. The purpose is not simply identifying the person who made the last mistake. It is understanding the organisational environment that allowed the mistake to pass through every defensive layer.
The objective is prevention rather than moral satisfaction.
That distinction matters because public anger naturally seeks punishment after tragedy.
Fifty-seven people are dead, many of them young.
Families want someone to answer for that.
Punishment may be justified where evidence establishes culpability.
But the more important public-policy objective is ensuring another fifty-seven people do not die the same way.
Those objectives overlap.
They are not identical.
A society can punish yesterday’s offender without fixing tomorrow’s railway.
There is also a privatisation question emerging because passenger services are operated by Hellenic Train, owned by Italy’s state-controlled Ferrovie dello Stato group, while rail infrastructure remains under Greek public-sector responsibility.
This can encourage another oversimplified argument: either privatisation caused the disaster or public-sector incompetence caused it.
The more useful question is how responsibilities were divided.
Fragmented systems can create interface risk.
One organisation operates trains.
Another controls infrastructure.
Another handles signalling.
Another regulates safety.
Each institution may perform its own task while coordination between them fails.
Privatisation is therefore neither automatically dangerous nor automatically efficient.
Its performance depends heavily on contract design, regulatory capacity and clarity over responsibility.
A private operator can be highly effective under strong regulation.
A public monopoly can be highly effective under strong management.
Either can fail under weak institutions.
Ownership is one variable.
Governance is another.
This matters because ideological arguments often jump directly from disaster to preferred economic model.
The left says privatisation failed.
The right says state bureaucracy failed.
Both may identify pieces of the causal structure.
Neither should be allowed to substitute ideology for investigation.
Evidence needs to determine which organisational failures actually mattered.
Prime Minister Kyriakos Mitsotakis initially described the crash largely through human error before later apologising and acknowledging that responsibility could not simply be placed on the station master. That shift itself demonstrates why language matters during institutional crises.
Leaders want to provide explanations quickly because uncertainty looks like incompetence.
Quick explanations can become politically dangerous when they appear to minimise systemic responsibility.
Sometimes the most credible sentence a government can say immediately after catastrophe is:
we do not know yet.
That is not weakness.
It is epistemic discipline.
The public does not need a complete causal story within twelve hours.
It needs confidence that evidence will be preserved, investigations will be independent and uncomfortable conclusions will be allowed to emerge.
Trust comes from the process.
Again, institutional credibility is accumulated through accepting the possibility that the institution itself may be responsible.
This is why independent accident investigation matters so much.
An agency investigating its own failures faces obvious conflicts of interest.
The state needs mechanisms capable of producing information the government may not want to hear.
That principle connects unexpectedly with China’s zero-COVID information problem. Very different political systems can suffer from the same organisational pathology: information becomes distorted when the people producing it fear the consequences of admitting failure.
A railway employee who has warned repeatedly about unsafe conditions may eventually stop reporting if nothing changes.
A manager may avoid escalating problems because doing so threatens budgets or careers.
A ministry may prefer optimistic progress reports.
Every level filters bad news.
The leadership then discovers the true condition of the system only when physical reality produces evidence too large to suppress.
Two trains collide.
This is why whistleblowers and unions can function as information institutions rather than merely interest groups.
They may possess their own agendas.
Their warnings should not automatically be accepted.
They should not automatically be dismissed either.
Organisations need channels through which uncomfortable information can move upwards without destroying the career of whoever delivers it.
Otherwise management eventually receives only good news.
Reality does not.
The European Union also belongs in the analysis. Greece has received European funding for infrastructure modernisation, and railway safety operates within EU regulatory frameworks. Later investigations would identify major delays in implementing EU-required control and communication systems and criticism of fragmented oversight.
European integration therefore does not automatically create European institutional quality.
Brussels can establish standards.
It can provide money.
National and local institutions still have to implement them.
This is another example of the gap between formal convergence and administrative convergence.
A country can adopt the same law as Germany without possessing Germany’s enforcement capacity.
EU membership changes rules faster than it changes institutions.
Institutional development takes time.
Greece has undergone enormous economic and political transformation since the debt crisis. Growth has returned and fiscal conditions have improved substantially from the darkest years of the 2010s.
That progress is real.
The railway crash is a reminder that macroeconomic recovery and institutional recovery are different processes.
GDP can grow before infrastructure is repaired.
Bond yields can fall before administrative capacity improves.
Investors can regain confidence before citizens regain confidence.
Economic indicators measure important things.
They do not measure everything governments are supposed to provide.
A functioning railway system is part of development too.
Perhaps that is the broader lesson from Tempi.
We often imagine state failure through spectacular images: coups, wars, governments collapsing.
Sometimes state failure is quieter.
A safety system that was supposed to be installed years ago.
A staffing vacancy.
A warning ignored.
A procurement delay.
A manual procedure continuing longer than intended.
Nothing dramatic happens.
Until everything happens at once.
The final collision may take seconds.
The institutional collision can take years.
That is why responsibility cannot stop with the person who happened to be standing beside the switch when the consequences finally arrived.
Human error is inevitable.
Catastrophe is what happens when institutions fail to expect it.
