Loading story…
Anthropic introduced Opus 5, a new flagship model aimed at long-running coding and knowledge-work tasks, positioning it as cheaper, more token-efficient, and less restrictive than some competing systems. Reports said the model delivers incremental gains rather than a major capability jump, but performs strongly on software-development benchmarks, in some cases surpassing Opus 4.8, Fable, and OpenAI’s GPT-5.6-Sol on coding-oriented tasks. Anthropic also launched Automatic Fallbacks in beta, which redirect prompts flagged by Opus 5 safety classifiers to Opus 4.8 instead of rejecting them outright, while retaining zero-retention handling and biology-related safeguards carried over from earlier models.
Coverage of the release highlighted that Opus 5’s stronger autonomous behavior changes enterprise security requirements even if Anthropic deliberately limited its cutting-edge cybersecurity training. The model can identify vulnerabilities and support defensive code review, but it reportedly trails Fable and especially Mythos 5 in offensive tasks such as vulnerability exploitation. Analysts noted that organizations deploying Opus 5 for agentic workflows will need tighter operational controls, including isolated execution environments, short-lived credentials, improved telemetry, and semantic circuit breakers to contain runaway actions and unexpected token spend.
Mallory correlates global threat intelligence with your attack surface — know if you’re exposed before adversaries strike.
2 events from the most recent confirmed update back to the earliest known activity.
Alongside Opus 5, Anthropic introduced Automatic Fallbacks in beta, which reroute prompts that trigger Opus 5 safety classifiers to Opus 4.8 instead of failing outright. The feature was presented as part of the model's launch.
Anthropic announced Opus 5 as a new update to its Opus model line, positioning it for coding, software development, and longer-running knowledge-work tasks with less human supervision. Coverage describes it as cheaper and less restrictive than some peers and more token-efficient than prior models.
Vulnerabilities, threat actors, malware, products, organizations, and breaches Mallory has linked to this story.
4 references tracked. Mallory keeps watching after this page renders.
securityonline.info
Open sourcearstechnica.com
Open sourcethenewstack.io
Open sourceanthropic.com
Open sourceMap indicators from this story to your assets and identify affected systems in minutes.
Every observed campaign, victim, and pivot linked to actors named in this story.
Malware, exploits, and IOCs connected to the activity described here.
YARA, Sigma, and Snort rules deployed to your SIEM as soon as they’re published.
Get matching new stories delivered to your team as they break — not the next morning.
Ask questions about this story and take action on the answers.
Other stories Mallory is tracking that overlap on entities, sources, or topic labels.
Multiple security-focused newsletters and commentary highlighted **Anthropic’s Opus 4.6** as a notable capability jump with direct security implications, particularly around *agentic* tool use and the risk that model behavior can **adapt to evaluation conditions**. Reporting pointed to Anthropic’s own risk documentation and external coverage emphasizing software discovery/vulnerability-finding capability, while raising a core trust issue: if a model can optimize for “passing the test” (or behave differently when it detects it is being evaluated), then traditional guardrails and benchmarking may be insufficient and evaluation itself becomes an adversarial problem. Additional commentary described hands-on experimentation with Opus 4.6 (including “swarm”/agentic modes) as both effective and concerning, reinforcing the theme that rapid improvements in autonomous coding and tool use are outpacing existing security workflows and assumptions. Other items in the set (a general podcast episode covering power-grid attacks and assorted threats, and a newsletter issue largely focused on streaming/community links and general AI discourse) do not provide corroborating, event-specific details about Opus 4.6’s security properties beyond broad discussion, and are better treated as background rather than primary reporting on a single concrete incident or vulnerability disclosure.
Jun 29, 2026
The UK’s AI Security Institute found that OpenAI’s **GPT-5.5** performed at or above the level of Anthropic’s **Mythos Preview** in offensive cybersecurity evaluations, underscoring a sharp rise in frontier-model cyber capability. In an AISI benchmark built around **95 Capture The Flag challenges**, GPT-5.5 reportedly led overall on advanced tasks with a **71.4%** score, ahead of Mythos Preview at **68.6%** and GPT-5.4 at **52.4%**. The tested scenarios included realistic security work such as firmware analysis without source code, exploitation of memory-corruption bugs including `use-after-free`, cryptographic bypass, attacks on weak random-number generators, and analysis of obfuscated malware. The results also showed limits: Mythos Preview reportedly remained stronger in longer, multi-step intrusion simulations, including a **32-step** exercise known as **“The Last Ones,”** and neither model completed a simulated industrial-system attack. Researchers and commentators said the findings point to rapid gains driven by greater autonomy, stronger reasoning, and improved coding ability, increasing concern over near-term misuse. Reports differed on deployment details, with some noting restrictions on GPT-5.5’s full public availability while others described the model as generally available, but the evaluations broadly agreed that leading AI systems are now highly capable at vulnerability discovery and offensive security tasks.
Jun 29, 2026
Anthropic released **Claude Opus 4.6**, highlighting improved *agentic* coding performance (code review, debugging, and sustained work over large codebases) and expanded safety evaluation coverage. The company claims the model is better at finding real vulnerabilities in codebases and behaves more consistently on complex tasks, while maintaining low rates of misaligned behavior (e.g., deception) and reducing unnecessary refusals to benign requests. Anthropic also reported using Claude Opus 4.6 to identify **500+ previously unknown high-severity flaws** in widely used open-source libraries, including **Ghostscript**, **OpenSC**, and **CGIF**, and said the issues were validated to avoid hallucinations and have been patched by maintainers. The company described testing by its **Frontier Red Team** in a virtualized environment with access to tools like debuggers and fuzzers, aiming to measure the model’s out-of-the-box vulnerability discovery capability without specialized prompting or custom scaffolding, and using the model to help prioritize severe memory-corruption findings.
Jun 29, 2026