The Source Signals
Tech news is everywhere. Intelligence isn't.
BeyondTrust Gains Access to Anthropic's Restricted Frontier Model to Hunt Flaws in Its Own Code
The identity security vendor joins Project Glasswing, turning a model Anthropic withholds from the public against the vulnerabilities in software that underpins governments and essential services.
AI security has a detection problem, and Check Point’s 2026 report puts a number on it
Only 13% of organisations can block a malicious prompt and just 5% can stop unsafe AI outputs, according to Check Point’s 2026 Cloud Security Report. Why detection has outpaced prevention.
The 2026 World Cup Is the Most Predictable Cyberattack Window Ever Handed to Adversaries
A Palo Alto Networks assessment finds disruptive intrusions, fraud at scale and hacktivist operations against the tournament are highly likely, with an active Iran-nexus campaign already targeting the municipal infrastructure 16 host cities will run under tournament load.
The 1% Problem: Inside Qualys’s bet that most of cybersecurity has been chasing the wrong list
This article is part of an editorial series produced in partnership with Qualys, exploring how the Risk Operations Center, is reshaping the way enterprises think about cyber risk. The reporting, interviews, and views are The Source Code’s own.
The biggest AI risk inside enterprises is the person at the keyboard, not the model, Optro report
With 82% of organisations reporting more AI-enabled attacks in the past year, new Optro research argues the dominant threat is no longer technical model failure but everyday human decisions made inside ungoverned AI tools.
From Backup to AI Trust: What Veeam Announced at VeeamON New York and Why It Matters
Three announcements in three days: a new platform built from the Securiti AI acquisition, a preview of Data Platform v13.1, and a governance framework exposing the gap between AI confidence and AI readiness.
Why Acronis Is Betting That the Next Decade of Managed Services Gets Built at the Infrastructure Layer, Not Bolted On to It
Cyber Frame, Cyber Console, Cyber Intelligence, Cyber Studio, and a rebuilt partner programme announced in a single broadcast signal that Acronis is no longer positioning itself as a backup and security vendor. It is positioning itself as the platform MSPs run their entire business on.
The AI Labs Built the Threat. Now They Are Selling the Defence
OpenAI's Daybreak launch brings a second AI lab into a cybersecurity race that Anthropic opened with Claude Mythos five weeks ago. For the security industry, the contest raises questions that go well beyond which model performs better.
275 Million Students, One Point of Failure, and No Way Out
The Canvas ransomware attack exposed what happens when a platform becomes too dominant to avoid and too architecturally fragile to trust — and students had no choice about being in either position.
AI security auditors can be fooled by code they cannot see past
Cloudflare’s threat research unit finds that volume, not deception, is the most reliable way to defeat an AI-powered security auditor.
1,200 arrests, 11,400 takedowns, one report: what Fortinet's 2025 sustainability disclosure actually says
From dismantling criminal networks across 18 nations to training nearly a million people in cybersecurity, Fortinet's 2025 Sustainability Report makes the case that protecting critical digital infrastructure is a public obligation — not just a product category.
China-aligned hackers exposed after leaving Slack and Discord logs intact during Mongolia espionage campaign
Cybersecurity firm ESET recovered thousands of attacker messages and draft emails after the previously undocumented GopherWhisper group used legitimate workplace tools as command-and-control infrastructure, then failed to clear its tracks.
Why Real-Time Cybersecurity Is Now the Operating System of Global Commercial Trust
An audit report ages the moment it is issued. A certificate begins decaying before it leaves the printer. In a market where AI agents act in milliseconds and insurers price detection time to the second, the periodic mechanisms that once produced trust have stopped working. The companies winning in the next decade are the ones rebuilding their security stacks to continuously demonstrate trust to audiences that never stop watching.
The machines are already inside: Microsoft's critical flaws doubled as AI agents flooded the enterprise
Critical Microsoft vulnerabilities doubled in 2025 while enterprise AI agents surged 467%, and the disclosure systems businesses rely on are losing the ability to see what matters most
The 500 strangers in your office that nobody hired
The rise of AI agents is introducing new challenges to traditional identity and access management, especially in identity registration, governance, and credential authentication. As autonomous agents access enterprise systems on behalf of humans, nobody can agree on who — or what — is responsible when something goes wrong. A reported feature on accountability in the age of machine identity.
The Mythos Reckoning: How One AI Model Is Rewriting the Rules of Cybersecurity
Anthropic's Claude Mythos Preview can autonomously find and exploit software vulnerabilities at a depth no human team can match. The industry response has ranged from alarm to pragmatism — and the most useful voices may be the ones urging both at once.
Most Organisations Cannot Actually Recover Data from a Cyberattack. They Just Think They Can
A new global report from Veeam exposes a widening gulf between recovery confidence and recovery reality, just as AI is accelerating the risks that make both matter more.
Under Attack and Still Standing: Building Cyber Resilience in a Geopolitical Crisis
Every geopolitical crisis brings a second offensive, one that runs not on battlefields but through networks, hospitals, water systems and businesses. Since the US-Iran war began in February 2026, that reality has moved from warning to weekly headline. Three cybersecurity leaders explain what it takes to stay standing when conflict follows you home through your own systems.
The Trust Problem: ChatGPT's Security Failures and What They Mean for Everyone Using It
Stolen GitHub credentials. Silent data exfiltration through DNS. Ungoverned AI agents roaming enterprise networks with administrator-level access. Three research disclosures, one sobering conclusion: ChatGPT has been handling the world's most sensitive data inside an environment that attackers could have exploited at will.
How a Single ChatGPT Prompt Could Silently Steal Your Data — and What Check Point Research Found Inside the Runtime
A hidden communication path inside ChatGPT's code execution environment let attackers drain sensitive conversations to external servers without triggering a single alert. The vulnerability is patched. But the security architecture question it raises is not.