Maple Grove Report

Maple Grove Report

Subscribe to Our Newsletter

Get our latest articles delivered straight to your inbox. No spam, we promise.


Report: Anthropic Deploys Engineers to Support NSA Use of Mythos

Pierluigi Paganini
June 06, 2026

Reports claim Anthropic engineers are helping the NSA use its restricted AI model Mythos, known for advanced cybersecurity capabilities.

This week, the Financial Times reported that Anthropic has placed approximately six “forward-deployed” engineers inside the National Security Agency to help the intelligence agency use Mythos, its most capable cyber model, for offensive operations. Two people familiar with the arrangement told the FT it would be useful for infiltrating networks in countries like China or Iran. Whether those engineers are involved in live operations, or only in customization and setup, remains unclear.

The reported collaboration comes amid tensions between Anthropic and the U.S. government. The company is challenging Pentagon policies over military use of AI, and was labeled a “supply-chain risk” after refusing to allow its models to be used for mass surveillance or autonomous weapons programs.

That designation triggered a procurement ban across the entire DoD, and President Trump has ordered the Pentagon to drop Claude from its systems by August. The NSA arrangement exists because of an explicit carve-out from that ban.

In April, Axios reported that the NSA was already using Mythos Preview.

“The National Security Agency is using Anthropic’s most powerful model yet, Mythos Preview, despite top officials at the Department of Defense — which oversees the NSA — insisting the company is a “supply chain risk,” two sources tell Axios.”

Anthropic’s public position on Mythos has been consistent: the model is too dangerous to release broadly because of what it can do. In an April 7 post, Anthropic’s red team said Mythos Preview could find and exploit zero-day vulnerabilities in “every major operating system and every major web browser” when a user directed it to. A complete exploit pipeline against a complex Linux target ran under a day at a cost below $2,000. Britain’s AI Security Institute, testing independently, found it solved 73% of expert-level tasks that no prior model could complete, and became the first model to finish a 32-step simulated corporate-network attack. That’s why access was limited to roughly 50 vetted organizations under Project Glasswing.

The company’s defenders have a ready answer for the NSA contradiction.

“The best way to build a good defence is to build a good attack.” reports the Financial Times.

A person close to the company told the FT, arguing that adversaries will build their own attack agents regardless. That’s not a bad argument on its merits. It also doesn’t explain why the same model deemed too risky for a security researcher to download is fine to deploy at a signals intelligence agency without public disclosure, legal framework, or congressional oversight anyone has confirmed.

The legal dispute is complicated. In May, a federal appeals court reviewed whether the Pentagon was justified in labeling Anthropic a “supply-chain risk.” One of the judges said she had not seen evidence that Anthropic acted with bad intent, despite Pentagon claims suggesting otherwise. However, the court appeared inclined to keep the designation in place. At the same time, Anthropic engineers were reportedly working with the NSA on Mythos, even though the agency falls under the Pentagon’s oversight.

On June 2, Anthropic expanded Project Glasswing from roughly 50 to approximately 150 organizations across more than 15 countries. The new cohort includes Okta, Samsung, NATO, and the EU’s cybersecurity agency ENISA. Partners have surfaced more than 10,000 high- or critical-severity flaws, and an internal Anthropic scan of 1,000 open-source projects flagged 23,019 potential vulnerabilities, 6,202 of them estimated high or critical. The expansion also added Australia’s Signals Directorate, extending government access beyond the US and UK for the first time.

Not everyone thinks the Mythos alarm was warranted to begin with. Reuters reported in May that fears about the model were overstated, quoting a researcher with early access who said vulnerability-hunting AI had been available for “months if not years.” The controlled rollout didn’t keep the capability rare; it just decided who holds it. Those holders now include a spy agency, a military alliance, and a chip manufacturer, days after Anthropic filed confidentially for an IPO at a valuation near $1 trillion, with annualized revenue on track to hit $50 billion by the end of June.

The NSA declined to confirm or deny the FT’s reporting. Anthropic didn’t respond to a request for comment from TechCrunch. The refusals, as the FT noted, are selective.

Follow me on Twitter: @securityaffairs and Facebook and Mastodon

Pierluigi Paganini

(SecurityAffairs – hacking, NSA)







Source link



TL;DR

Hackers breached five Polish water treatment plants using default passwords and internet-exposed control systems. Poland is spending a billion euros on cybersecurity; 70 per cent of American water utilities fail the same basic standards.

Hackers breached five Polish water treatment plants in 2025, gaining access to the industrial control systems that regulate pumps, filters, and chemical dosing. In some facilities, the attackers could have altered the operational parameters of equipment that determines what comes out of the tap. The attack vector, in every case, was unremarkable: weak passwords and control systems connected directly to the internet.

Poland’s Internal Security Agency, the ABW, disclosed the breaches this week in its first public activity summary since 2014, before Russia annexed Crimea. The report names the facilities: JabÅ‚onna Lacka, Szczytno, MaÅ‚dyty, Tolkmicko, and Sierakowo, five small towns whose water treatment stations were found to have been penetrated by attackers the agency attributes, with careful phrasing, to “hacktivist groups” that are “often personas used by foreign governments, particularly Russian intelligence services.

The breaches

The incidents were not theoretical. In Szczytno, in May 2025, someone accessed the supervisory control system and changed flushing cycles while the facility was being monitored on a live feed. In JabÅ‚onna Lacka, in September, a video captured an intruder logging in through an admin account and manipulating pump and filter thresholds. The ABW said the attackers had the ability to alter technical parameters of devices, creating “a direct risk” to the continuity of water supply operations.

The agency identified two primary attack vectors: passwords that had not been changed from factory defaults and industrial control systems exposed directly to the public internet. Neither vulnerability requires sophisticated tooling to exploit. Both have been documented in cybersecurity advisories for more than a decade.

The ABW report names Russian APT groups including APT28 and APT29, and the Belarusian-linked group UNC1151, as operating against Polish targets. The agency stopped short of attributing specific water treatment breaches to specific groups, but the pattern is consistent with a broader escalation that Poland’s government says has made the country the target of between 20 and 50 cyberattacks per day.

The escalation

Cyberattacks on Poland surged after the election of its pro-Ukraine government, and the tempo has not slowed. In December 2025, a coordinated attack hit a combined heat and power plant supplying heat to almost 500,000 customers, along with multiple wind and solar farms. The cybersecurity firm ESET attributed the attack to Sandworm, a group the United States government has linked to Russia’s military intelligence directorate, the GRU.

Poland’s cybersecurity budget for 2026 is a record one billion euros, up from 600 million in 2024. Of that, 80 million euros has been allocated specifically to the cyber defences of water management systems. Germany has absorbed 90 per cent of Europe’s record defence tech funding, but Poland’s per-capita spending on cybersecurity now exceeds that of most NATO members.

The spending reflects a recognition that the threat has moved beyond espionage. Helsing, the European military AI startup, raised 450 million euros explicitly to defend NATO from Russia, and Ukraine’s emergence as a defence tech powerhouse has demonstrated that the countries closest to Russia’s borders are now building the capabilities to respond. But the water treatment plants in JabÅ‚onna Lacka and Szczytno were not breached by advanced persistent threats deploying novel exploits. They were breached because someone left the default password on a system connected to the internet.

The American parallel

The United States faces the same vulnerability at a larger scale. In 2024, the Environmental Protection Agency found that nearly 70 per cent of water utilities inspected by federal officials were in violation of basic cybersecurity standards, including the failure to change default passwords. The largest regulated water and wastewater utility in the country, American Water, was forced to shut down its billing systems in October 2024 after a cyberattack disrupted services for millions of customers.

The threats are not hypothetical. The Chinese state-sponsored group Volt Typhoon has compromised the information technology environments of multiple US critical infrastructure organisations, including water and wastewater systems, in what CISA, the NSA, and the FBI assess is an effort to pre-position for disruptive or destructive cyberattacks in the event of a major crisis or conflict. The Iranian-affiliated group CyberAv3ngers has targeted programmable logic controllers at US water treatment plants, including facilities in Pennsylvania.

The EPA, CISA, and the FBI have issued repeated advisories. Congress temporarily reinstated cybersecurity information-sharing authorities in November 2025, then let them lapse again in January 2026. The federal government has published cybersecurity planning tools, incident response templates, and procurement checklists. The water utilities that need them most are the ones least likely to use them: small municipal systems with limited budgets, ageing infrastructure, and no dedicated cybersecurity staff.

The gap

Defence stocks are surging across Europe as governments pour money into military technology. Poland is spending a billion euros on cybersecurity. NATO is funding innovation accelerators and defence AI alliances. The investment reflects an accurate assessment of the threat.

But the water treatment plants that were breached in Poland were not protected by any of it. The facilities in Jabłonna Lacka and Szczytno were running control systems with factory-default credentials exposed to the internet. The American utilities that the EPA found in violation of basic standards are running the same configuration. The sophistication of the attacker is irrelevant when the front door is unlocked.

Poland’s ABW published its first activity summary in a decade because the scale of the threat has made silence untenable. The United States has published advisory after advisory. The pattern is consistent across both countries: the governments that understand the threat best are the ones whose critical infrastructure remains most exposed, because the systems that treat drinking water are operated by municipalities that lack the resources, the expertise, or the regulatory compulsion to secure them. The hackers who breached five Polish water plants did not need a zero-day exploit. They needed a password.



Source link

Recent Reviews