8-minute read | 1,850 words
What to know this week
Google announces hackers are using AI to create zero days.
Google’s threat intelligence group found that attackers are using artificial intelligence (AI) to create new zero-day exploits.
Canvas has paid hackers for the return of stolen data.
Canvas, one of the largest learning platform providers, has agreed to pay hackers for the return of its stolen data after suffering a major breach.
This week's full stories
Google researchers find that threat actors are utilizing AI.
THE NEWS
On Monday, Google’s Threat Intelligence Group (GTIG) released a new report detailing how threat actors are utilizing AI systems. In the report, Google’s researchers noted “a maturing transition from nascent AI-enabled operations to the industrial-scale application of generative models within adversarial workflows.” Furthermore, GTIG researchers focused on:
- Vulnerability Discovery and Exploit Generation.
- AI-Augmented Development for Defense Evasion.
- Autonomous Malware Operations.
- AI-Augmented Research and IO.
- Obfuscated LLM Access.
- Supply Chain Attacks.
From their research, this is the first time researchers have found that AI was being used to develop these types of vulnerabilities.
John Hultquist, chief analyst at GTIG, noted:
“For every zero-day we can trace back to AI, there are probably many more out there. Threat actors are using AI to boost the speed, scale, and sophistication of their attacks.”
Alongside noting that AI is being used to create new zero-day exploits, GTIG researchers also found state-sponsored threat actors have begun to utilize these tools and models. For example, the report found that Russian-linked hacking groups have been using AI models to target Ukrainian networks with malware alongside North Korean groups that are using it to refine and scale their cyber methods.
THE KNOWLEDGE
Over recent weeks, AI’s impact and involvement in cybersecurity has become increasingly scrutinized. With the launches of Anthropic’s Claude Mythos Preview and similar models, there have been growing concerns that these advanced cyber deception models could be utilized by threat actors to expand the scope of their attack capabilities.
These concerns are not new. For months, researchers have continued to find that threat actors have already begun to enhance their attacks. For example, Anthropic researchers found that Chinese state-sponsored groups were already manipulating previous iterations of Claude Code tools to aid in infiltration attempts. More specifically, researchers noted that these operations involved infiltration attempts at roughly thirty global targets with reported success in a small number of cases. These targets included large tech companies, financial institutions, government agencies, and chemical manufacturers.
As newer, more sophisticated models continue to emerge, there are growing concerns within the government about the potential for misuse. The Trump administration has already begun to consider pivoting its established AI strategy. At the beginning of May, reports emerged that the Trump administration was considering establishing a new oversight committee to review all new AI models before they could be released to the general public.
As the administration continues to weigh its oversight options, reports similar to GTIG’s and those evaluating Claude Mythos will likely be significant when determining what policies are needed to properly manage these industry-defining AI models.
THE IMPACT
While AI models hold significant cybersecurity promise, potentially reducing costs and improving response efforts, there are significant concerns that these models will do just as much harm when utilized by threat actors. These systems are no longer simply improving phishing attempts or generating basic malware code. Rather, GTIG’s findings show that adversaries are using these capabilities to enhance all of their offensive lifecycle operations including vulnerability research, exploit development, planning, and evasion.
For policymakers, this creates a major strategic dilemma. If AI meaningfully accelerates attack capabilities, governments will continue to face greater pressure to regulate these tools. However, overregulating these tools could hamper innovation and cause significant delays in new product releases.
Nonetheless, this report will continue to add to growing pressures that AI tools need stronger governance. That increasing pressure could result in new oversight frameworks, pre-release testing requirements, and reporting mandates for AI developers. If AI compresses the time between vulnerability discovery and exploitation, the current patch-and-defend model will also likely continue to become increasingly unsustainable.
Canvas pays hackers.
THE NEWS
On Tuesday, Canvas' parent company, Instructure, announced that it would be paying hackers who recently stole the company's data in an attack, alongside paying for the attackers to delete the data copies. Additionally, Instructure assured Canvas users that no one would face extortion from the attack.
In a statement, the company emphasized:
“While there is never complete certainty when dealing with cybercriminals, we believe it was important to take every step within our control to give customers additional peace of mind, to the extent possible.”
For context, Canvas, one of the United States (US) largest online education platforms, was hacked last week by the infamous hacking group, ShinyHunters. In the attack, ShinyHunters were able to take Canvas's platform offline, alongside having access to over 275 million student records across 9,000 institutions. These records included full names, message contents, and email addresses.
After the initial attack, the group threatened to begin publishing the stolen data if a ransom was not paid. This data included “several billions of private messages among students and teachers.”
Instructure has not commented on how much they paid ShinyHunters.
THE KNOWLEDGE
As threat tactics have continued to evolve and change, ransomware has continued to remain one of the most effective tools for attackers. However, ransomware has also begun to see some unique trends. When examining ransomware trends, Bitsight found that both double extortion tactics and Ransomware-as-a-Service (RaaS) are likely to continue. Additionally, they found that the US experienced the most ransomware attacks by a large percentage accounting for 3987 attacks or 36.6% of the total in the past year.
For greater detail, double extortion ransomware significantly increases pressure on its victims. In this tactic, threat actors not only encrypt the data they steal but then also threaten to leak it, as seen with the Canvas’s case. This tactic first emerged as a noticeable trend in 2019 through Maze and REvil ransomware strains. When utilizing double extortion, cybercriminals will both encrypt the data and steal it. This strategy helps get around defensive back-ups as now attackers can also threaten data exposure if a ransom is not paid.
RaaS is an emerging cybercrime business model where actors will sell ransomware code or malware to hackers. Through this model, hackers are able to profit from extortion without needing to develop their own malware. On the other hand, ransomware creators are able to increase their profits without having to launch attacks of their own. Sold services can involve monthly subscriptions, one-time fees, affiliate programs, and profit-sharing models. The emergence of RaaS makes it harder to attribute attacks, makes ransomware threats more resilient, and creates new pressure tactics.
THE IMPACT
Instructure’s decision to pay ShinyHunters reflects a growing reality across both the public and private sectors. While Instructure framed the payment as a step to protect customers, the decision reinforces a dangerous precedent that cyber extortion works, especially against those who possess sensitive personal data.
More broadly, the breach highlights how ransomware groups are shifting away from targeting high-value enterprises alone and now also focusing on vendors upstream of thousands of institutions. This strategy could allow attackers to gain leverage over a much larger network, increasing how much data is harvested and how much they could potentially extort.
The Canvas breach shows that ransomware is no longer just a corporate risk but rather now an ecosystem risk. As attackers potentially pivot to target shared digital platforms, the downstream impacts can span thousands of institutions at once. This raises questions on if providers will need to face stronger cybersecurity requirements or risk management procedures.
This Week's Caveat Podcast: Sony's failed attempt to stop piracy.
Dave Bittner and Ben Yelin sat down to discuss how Sony’s failed lawsuit against ISPs could impact other copyright lawsuits and another story regarding how the EU’s approach to AI might be informed by nuclear deterrence strategies. Afterwards, we sat down with Dr. Liz James, a managing security consultant at NCC Group. Throughout the conversation, Dr. James discusses a series of new cybersecurity regulations that were instituted through the EU’s vehicle emissions regulations.
OTHER NOTEWORTHY STORIES
EU Commission in talks with OpenAI and Anthropic over models.
What: The European Commission is in ongoing talks with OpenAI and Anthropic regarding access to new AI models.
Why: On Monday, reports emerged that the European Commission is engaging with both major US AI firms regarding securing access to new frontier AI models.
Spokesperson Thomas Regnier commented on the meetings, stating:
“With one (OpenAI), you have a company proactively offering to give access to the company. With the other one (Anthropic), we have good exchanges though we’re not at a stage where we can speculate on potential access or not.”
Currently, the Commission has had several meetings with Anthropic though none of these meetings involved possible access to its AI models.
MAY 11, 2026 | Source: Reuters
Senate Committee set to consider crypto bill.
What: US Senators are set to consider a long-awaited cryptocurrency bill.
Why: On Friday, Senators announced they would consider a cryptocurrency bill known as the Clarity Act. The act, if signed, would aim to provide greater clarity on financial regulators’ jurisdiction over the emerging currencies.
Senator Tim Scott, the chairman of the Senate Banking Committee, stated that the committee would hold an executive session on May 14, 2026.
Currently, the bill also holds a key provision. In the section, customer rewards on idle holdings of stablecoins would be prohibited.
MAY 8, 2026 | Source: Reuters
Netflix sued by Texas for spying on minors.
What: Texas has filed a lawsuit against Netflix alleging that the company was spying on minors by collecting their data without consent.
Why: On Monday, Texas Attorney General Ken Paxton filed a lawsuit against Netflix. In the suit, Texas alleges that Netflix has falsely represented to consumers that it did not collect or share user data for years, when in reality, the company tracked and sold viewer habits and preferences to data brokers.
According to Texas’s complaint:
“Netflix’s endgame is simple and lucrative: get children and families glued to the screen, harvest their data while they are stuck there, and then monetize the data for a handsome profit.”
MAY 11, 2026 | Source: Reuters
BrentWorks launches new tool to catch AI hallucinations in legal citations.
What: BrentWorks Inc launched CiteSentinel, a new tool dedicated to preventing AI hallucinations in legal citations.
Why: Last week, BrentWorks, a legal tech startup, launched its new tool CiteSentinel. The tool is designed to detect AI hallucinations and prevent them in legal documents. The tool operates by scanning documents and flags case law, statutes, and legal authorities that could be potentially fabricated, misstated, or erroneous before being submitted to a judge.
Brent Britton, the co-founder of BrentWorks, stated:
“Lawyers are learning that AI makes mistakes and lies confidently. CiteSentinel restores trust, helping lawyers use generative AI while filing documents grounded in reputable authorities.”
MAY 7, 2026 | Source: EIN News