The information that is emerging about Russia’s extensive cyberintelligence operation against the United States and other countries should be increasingly alarming to the public. The magnitude of the hacking, now believed to have affected more than 250 federal agencies and businesses — primarily through a malicious update of the SolarWinds network management software — may have slipped under most people’s radar during the holiday season, but its implications are stunning.
According to a Washington Post report, this is a massive intelligence coup by Russia’s foreign intelligence service (SVR). And a massive security failure on the part of the United States is also to blame. Our insecure Internet infrastructure has become a critical national security risk — one that we need to take seriously and spend money to reduce.
President-elect Joe Biden’s initial response spoke of retaliation, but there really isn’t much the United States can do beyond what it already does. Cyberespionage is business as usual among countries and governments, and the United States is aggressively offensive in this regard. We benefit from the lack of norms in this area and are unlikely to push back too hard because we don’t want to limit our own offensive actions.
Biden took a more realistic tone last week when he spoke of the need to improve US defenses. The initial focus will likely be on how to clean the hackers out of our networks, why the National Security Agency and US Cyber Command failed to detect this intrusion and whether the 2-year-old Cybersecurity and Infrastructure Security Agency has the resources necessary to defend the United States against attacks of this caliber. These are important discussions to have, but we also need to address the economic incentives that led to SolarWinds being breached and how that insecure software ended up in so many critical US government networks.
Software has become incredibly complicated. Most of us almost don’t know all of the software running on our laptops and what it’s doing. We don’t know where it’s connecting to on the Internet — not even which countries it’s connecting to — and what data it’s sending. We typically don’t know what third party libraries are in the software we install. We don’t know what software any of our cloud services are running. And we’re rarely alone in our ignorance. Finding all of this out is incredibly difficult.
This is even more true for software that runs our large government networks, or even the Internet backbone. Government software comes from large companies, small suppliers, open source projects and everything in between. Obscure software packages can have hidden vulnerabilities that affect the security of these networks, and sometimes the entire Internet. Russia’s SVR leveraged one of those vulnerabilities when it gained access to SolarWinds’ update server, tricking thousands of customers into downloading a malicious software update that gave the Russians access to those networks.
The fundamental problem is one of economic incentives. The market rewards quick development of products. It rewards new features. It rewards spying on customers and users: collecting and selling individual data. The market does not reward security, safety or transparency. It doesn’t reward reliability past a bare minimum, and it doesn’t reward resilience at all.
This is what happened at SolarWinds. A New York Times report noted the company ignored basic security practices. It moved software development to Eastern Europe, where Russia has more influence and could potentially subvert programmers, because it’s cheaper.
Short-term profit was seemingly prioritized over product security.
Companies have the right to make decisions like this. The real question is why the US government bought such shoddy software for its critical networks. This is a problem that Biden can fix, and he needs to do so immediately.
The United States needs to improve government software procurement. Software is now critical to national security. Any system for acquiring software needs to evaluate the security of the software and the security practices of the company, in detail, to ensure they are sufficient to meet the security needs of the network they’re being installed in. Procurement contracts need to include security controls of the software development process. They need security attestations on the part of the vendors, with substantial penalties for misrepresentation or failure to comply. The government needs detailed best practices for government and other companies.
Some of the groundwork for an approach like this has already been laid by the federal government, which has sponsored the development of a “Software Bill of Materials” that would set out a process for software makers to identify the components used to assemble their software.
This scrutiny can’t end with purchase. These security requirements need to be monitored throughout the software’s life cycle, along with what software is being used in government networks.
None of this is cheap, and we should be prepared to pay substantially more for secure software. But there’s a benefit to these practices. If the government evaluations are public, along with the list of companies that meet them, all network buyers can benefit from them. The US government acting purely in the realm of procurement can improve the security of nongovernmental networks worldwide.
This is important, but it isn’t enough. We need to set minimum safety and security standards for all software: from the code in that Internet of Things appliance you just bought to the code running our critical national infrastructure. It’s all one network, and a vulnerability in your refrigerator’s software can be used to attack the national power grid.
The IOT Cybersecurity Improvement Act, signed into law last month, is a start in this direction.
The Biden administration should prioritize minimum security standards for all software sold in the United States, not just to the government but to everyone. Long gone are the days when we can let the software industry decide how much emphasis to place on security. Software security is now a matter of personal safety: whether it’s ensuring your car isn’t hacked over the Internet or that the national power grid isn’t hacked by the Russians.
This regulation is the only way to force companies to provide safety and security features for customers — just as legislation was necessary to mandate food safety measures and require auto manufacturers to install life-saving features such as seat belts and air bags. Smart regulations that incentivize innovation create a market for security features. And they improve security for everyone.
It’s true that creating software in this sort of regulatory environment is more expensive. But if we truly value our personal and national security, we need to be prepared to pay for it.
The truth is that we’re already paying for it. Today, software companies increase their profits by secretly pushing risk onto their customers. We pay the cost of insecure personal computers, just as the government is now paying the cost to clean up after the SolarWinds hack. Fixing this requires both transparency and regulation. And while the industry will resist both, they are essential for national security in our increasingly computer-dependent worlds.
This essay previously appeared on CNN.com.
Researchers have been able to find all sorts of personal information within GPT-2. This information was part of the training data, and can be extracted with the right sorts of queries.
Abstract: It has become common to publish large (billion parameter) language models that have been trained on private datasets. This paper demonstrates that in such settings, an adversary can perform a training data extraction attack to recover individual training examples by querying the language model.
We demonstrate our attack on GPT-2, a language model trained on scrapes of the public Internet, and are able to extract hundreds of verbatim text sequences from the model’s training data. These extracted examples include (public) personally identifiable information (names, phone numbers, and email addresses), IRC conversations, code, and 128-bit UUIDs. Our attack is possible even though each of the above sequences are included in just one document in the training data.
We comprehensively evaluate our extraction attack to understand the factors that contribute to its success. For example, we find that larger models are more vulnerable than smaller models. We conclude by drawing lessons and discussing possible safeguards for training large language models.
We generated a total of 600,000 samples by querying GPT-2 with three different sampling strategies. Each sample contains 256 tokens, or roughly 200 words on average. Among these samples, we selected 1,800 samples with abnormally high likelihood for manual inspection. Out of the 1,800 samples, we found 604 that contain text which is reproduced verbatim from the training set.
The rest of the blog post discusses the types of data they found.
Wired has a detailed story about the ransomware attack on a Dusseldorf hospital, the one that resulted in an ambulance being redirected to a more distant hospital and the patient dying. The police wanted to prosecute the ransomware attackers for negligent homicide, but the details were more complicated:
After a detailed investigation involving consultations with medical professionals, an autopsy, and a minute-by-minute breakdown of events, Hartmann believes that the severity of the victim’s medical diagnosis at the time she was picked up was such that she would have died regardless of which hospital she had been admitted to. “The delay was of no relevance to the final outcome,” Hartmann says. “The medical condition was the sole cause of the death, and this is entirely independent from the cyberattack.” He likens it to hitting a dead body while driving: while you might be breaking the speed limit, you’re not responsible for the death.
So while this might not be an example of death by cyberattack, the article correctly notes that it’s only a matter of time:
But it’s only a matter of time, Hartmann believes, before ransomware does directly cause a death. “Where the patient is suffering from a slightly less severe condition, the attack could certainly be a decisive factor,” he says. “This is because the inability to receive treatment can have severe implications for those who require emergency services.” Success at bringing a charge might set an important precedent for future cases, thereby deepening the toolkit of prosecutors beyond the typical cybercrime statutes.
“The main hurdle will be one of proof,” Urban says. “Legal causation will be there as soon as the prosecution can prove that the person died earlier, even if it’s only a few hours, because of the hack, but this is never easy to prove.” With the Düsseldorf attack, it was not possible to establish that the victim could have survived much longer, but in general it’s “absolutely possible” that hackers could be found guilty of manslaughter, Urban argues.
And where causation is established, Hartmann points out that exposure for criminal prosecution stretches beyond the hackers. Instead, anyone who can be shown to have contributed to the hack may also be prosecuted, he says. In the Düsseldorf case, for example, his team was preparing to consider the culpability of the hospital’s IT staff. Could they have better defended the hospital by monitoring the network more closely, for instance?
Documented Death from a Ransomware Attack
A Dusseldorf woman died when a ransomware attack against a hospital forced her to be taken to a different hospital in another city.
I think this is the first documented case of a cyberattack causing a fatality. UK hospitals had to redirect patients during the 2017 WannaCry ransomware attack, but there were no documented fatalities from that event.
The police are treating this as a homicide.
Sidebar photo of Bruce Schneier by Joe MacInnis.
Attacks Hit Many UK Firms
- Government report says seven in ten large companies identified a breach or attack with firms holding personal data more likely to be attacked primarily by fraudulent emails, followed by viruses and
- Businesses large and small are being urged to protect themselves against cyber-crime after new government statistics found more than half of all UK businesses suffered a cyber-breach or attack in the past 12
- The Cyber-Security Breaches Survey 2017 reveals nearly seven in ten large businesses identified a breach or attack, with the average cost to large businesses of all breaches over the period being £20,000 and in some cases reaching
- The survey also shows businesses holding electronic personal data on customers were much more likely to suffer cyber-breaches than those that do not (51% compared to 37%)
- The Cyber-Breaches Survey is part of the UK government’s five-year national cyber-security strategy to transform this country’s cyber-security and to protect the UK online. As part of the strategy, the government recently opened the new National Cyber-Security Centre (NCSC), a part of
It’s the shortest month of the year, but it has not been short of breaches or cyber attacks.
Are You Covered For CyberAttack ?
BLOOMFIELD, Conn.—A recent study from NTT Com Security, found that 49 percent of the U.S. companies surveyed currently do not have insurance specifically for cybersecurity attacks.
NTT Com Security surveyed 1,000 “non-IT business decision makers in organizations in the U.K., U.S., Germany, France, Sweden, Norway and Switzerland,” for the report.
“Faced with risks every day, it’s easy for organizations to look for quick-fix solutions rather than focusing on building a solid security and risk management strategy,” Garry Sidaway, SVP security strategy and alliances for NTT Com Security, said in a prepared statement.
“Rather than relying solely on an insurance policy to cover losses, businesses need a different game plan. Buy insurance by all means, but ensure that you can demonstrate that you have put controls in place to reduce your risks, and, what these controls cover. This way you know what is being insured,” he said.
While a majority of global organizations believe information security breach insurance is crucial, less than half—41 percent—are fully covered for both security breaches and data loss, and just over one-third have dedicated cybersecurity insurance, according to the company’s 2016 Risk:Value report.
U.S. businesses are the most likely to have this type of insurance, 51 percent, compared to 26 percent in the U.K.
“Security needs to be embedded into the culture of an organization, from top to bottom, championed by the CEO, designed and executed by the CISO and communicated effectively so that every employee takes responsibility for ensuring that good practices are followed,”