Top news stories this week
- Rise of the machines. FBI, AWS and Google warn of rise in agentic cyberattacks.
- Shaky foundations. AI-found flaws hit Java and PHP as Optimizely hack poisons major websites.
- Caught in the web. Law enforcement closes in on cyber criminals.
- Sparks fly. Fireworks breach spreads across customer environments.
- Out of office? Microsoft outage highlights digital dependencies with real-world consequences.
- Agents of chaos. Autonomous agents continue testing limits of access controls, sandboxing, and human oversight.
1. FBI, AWS and Google warn of rise in agentic cyberattacks
The FBI warned that threat actors are weaponising AI-powered offensive frameworks such as Artex to dramatically accelerate cyberattacks. AWS observed Artex campaigns to steal AI credentials and bypass security controls at scale, while Google Threat Intelligence Group reported AI tools being used to automate vulnerability discovery and exploitation.
So what?
AI is changing the speed and scale of attacks more than the underlying techniques. Read 'S-RM lessons from responding to agentic AI attacks' and sign up to our webinar Understanding agentic AI-enabled cyberattacks to learn more.
[Researcher: Milda Petraityte]
2. AI-found flaws hit Java and PHP as Optimizely hack poisons major websites
Attackers exploited a flaw in Optimizely, a popular website-testing platform, to inject malicious code into sites belonging to 178 organisations, with the aim of stealing visitors' payment card details and cryptocurrency.
Separately, start-up Calif used AI to uncover "Java4Shell", a critical flaw in Java that can allow unauthenticated server takeover, prompting CISA to urge organisations to patch immediately. Calif previously helped find "PHP2Shell", two PHP flaws exploited to breach firewalls, while more than half of exposed servers remain unpatched. PHP has since fixed 18 further flaws, seven critical.
So what?
AI is speeding up how quickly flaws in widely used software are found and turned into working attacks. Organisations should urgently patch Java and PHP, starting with internet-facing systems, and review which third-party scripts their websites load.
[Researcher: Asya Sonnichsen]
3. Law enforcement closes in on cyber criminals
A British member of the Scattered Spider cybercriminal group has been sentenced to six years in prison in the US for their role in phishing attacks that compromised companies and led to the theft of more than USD 11.6 million in cryptocurrency.
Separately, a Kazakh national has been extradited from Italy to the US over alleged involvement with INC and Lynx ransomware operations targeting hospitals and schools, while Ukrainian authorities together with the FBI have arrested four suspects linked to more than USD 40.5 million in ransom payments.
So what?
These law enforcement operations demonstrate the growing reach of efforts to combat cybercriminals operating across borders. However, underlying threats are unlikely to dissipate with other affiliates and groups likely to fill any gaps left by disrupted actors.
[Researcher: Robin Toulalan Parr]
4. Fireworks breach spreads across customer environments
A breach at AI infrastructure provider Fireworks AI impacted multiple customers, including Uber, where attackers used compromised credentials to access personal information belonging to 149,346 drivers.
The breach also affected AI coding platform Cursor. Attackers used an exposed Fireworks AI key to access coding-agent traces and steal additional credentials, one of which enabled them to place trojanised versions of Anthropic's Claude Code extension in Cursor's marketplace cache, potentially reaching up to 5,000 users.
SO WHAT?
As organisations expand their AI supply chains, they should apply the same scrutiny to third-party access and credentials as other critical technology dependencies. Least-privilege access, effective secrets management and rapid credential rotation can help limit the impact of a supplier breach.
[Researcher: Gabriella Nolan]
5. Microsoft outage highlights digital dependencies with real-world consequences
A nearly 24-hour outage of Microsoft services including Outlook, Exchange Online and Teams, was triggered by a combination of a software bug, increased traffic and reduced server capacity, causing cascading failures across Microsoft’s infrastructure. In one instance, this disruption had real-world consequences for 15 guests at a German hotel, who experienced health problems after an email alert for excessive chlorine levels its pools failed to reach operators.
So What?
Cloud services can carry operational risks beyond temporary IT disruption, particularly where they support critical processes. Organisations should aim to identify critical dependencies and ensure appropriate backup communication or failover mechanisms are available if primary services become unavailable.
[Researcher: Jenny Eysert]
6. Autonomous agents continue testing limits of access controls, sandboxing, and human oversight
Researchers reported that Grok bots made dozens of unauthorised attempts to access corporate systems despite being instructed to stop, although the company's access gateway blocked them. Separately, the state of California is suing OpenAI over alleged failures to contain agents that escaped sandbox-style testing environments and attacked real-world systems. In addition, Google Gemini agents repeatedly bypassed safeguards during testing to probe a live website, attempt SQL injection, and impersonate employees.
SO WHAT?
Organisations must treat AI agents as privileged identities, not passive productivity tools. Practitioners are encouraged to restrict connected applications and permissions, require approval for sensitive actions, monitor agent activity, and enforce network-level isolation that cannot be overridden by the agent itself.
[Researcher: Steve Ross]
The information contained herein is intended for informational purposes only. Insurance coverage in any particular case will depend upon the type of policy in effect, the terms, conditions and exclusions in any such policy, and the facts of each unique situation. No representation is made that any specific insurance coverage would apply in the circumstances outlined herein. Please refer to the individual policy forms for specific coverage details. S-RM does not assume, and shall have no liability for the control, correction, continuation or modification of any existing conditions or situation.
